Feat: Add environment variables for path customization and centralize config

[bsirak]

Description:
This PR introduces the ability to customize core uvr directory paths via environment variables (similar to how uv handles them) and centralizes all environment variable reads into a single config.rs module for better maintainability and documentation.

New Features:
Added support for overriding previously hardcoded paths using the following new environment variables:

• UVR_CACHE_DIR: Customizes the directory where packages, environments, and tarballs are cached (default: ~/.uvr/cache/).
• UVR_R_VERSIONS_DIR: Customizes the directory where managed R versions are installed (default: ~/.uvr/r-versions/).
• UVR_INSTALL_DIR: Customizes the directory where the uvr binary itself is installed during a self-update.

Refactoring & Consolidation:
To keep configuration consistent, previously existing environment variables scattered throughout the codebase have been moved into crates/uvr-core/src/config.rs:

• UVR_EXTRA_LIBS (moved from run.rs)
• UVR_INSTALL_TIMEOUT (moved from r_cmd_install.rs)
• UVR_LIBRARY (moved from sync.rs)
• UVR_PROGRESS (moved from progress.rs)

Changes:

• Created crates/uvr-core/src/config.rs with alphabetically sorted accessor functions for all UVR_* variables.
• Added detailed docstrings for each variable, explaining expected formats and default behaviors.
• Updated all call sites across the codebase (e.g., manager.rs, downloader.rs, cache.rs, sync.rs, etc.) to use the new centralized getters instead of hardcoded paths or inline std::env::var checks.

[bsirak]

I kept UVR_R_VERSIONS_DIR because that's what we discussed, but I must say it bothers me it's not UVR_R_INSTALL_DIR for reasons...

[nbafrank]

Hey @bsirak ! Thanks so much for this! You are totally right about UVR_R_INSTALL_DIR because it is more standardized with both uv but also pairs with UVR_INSTALL_DIR, go for it and make that change!

I have a few things for you!

Naming — you're right!

Let's rename UVR_R_VERSIONS_DIR → UVR_R_INSTALL_DIR as discussed above!

CI — formatting only

CI is failing on cargo fmt --all -- --check across all three platforms. No compile or test failures — just rustfmt. Running cargo fmt
--all locally and pushing should clear it. Diffs are in cache.rs:8, doctor.rs:324, sync.rs:308/:675, and three spots in
self_update.rs (trailing whitespace, anyhow::bail! wrapping, inline-comment placement).

Blocking — correctness

crates/uvr-core/src/registry/cran.rs:346 — silent CRAN cache invalidation. The .join("cran") moves the CRAN package-index from
~/.uvr/cache/{key}-packages.txt to ~/.uvr/cache/cran/{key}-packages.txt. Every existing user's CRAN cache becomes orphaned on first
run after upgrade, and the layout is now inconsistent with P3M which stays flat at ~/.uvr/cache/p3m-*. Either drop the .join("cran")
here, or namespace P3M too — but pick one. I'd drop it; cache organization can be a separate PR if we want to do it intentionally for
both registries.

crates/uvr/src/commands/self_update.rs — UVR_INSTALL_DIR footgun. When UVR_INSTALL_DIR differs from current_exe.parent(), the new
binary is written to the override dir but the running binary at current_exe is never touched. Depending on PATH order, the next
invocation may hit the stale binary and the user thinks the update failed. Either reject the mismatch with a clear error
("UVR_INSTALL_DIR points to X but uvr is running from Y; remove the env var or move uvr to the override dir first"), or at minimum
emit a warn explaining the user must update PATH.

self_update.rs — lost error. The old code errored if current_exe.parent() was None; the new code silently falls back to
Path::new("."). Please restore the explicit error — silent install-into-CWD is worse than a clear failure.

Polish

• config.rs:32-37 — install_dir() has a TODO-style comment ("We could fall back to ~/.local/bin but if we are just reading an env var
  to override... maybe it just defaults to None... Let's provide a function that returns the env var or None.") — this reads like
  working notes. Either commit to the choice with a one-line rationale, or remove.
• run.rs minor — let extra = extra.trim().to_string(); introduces an unnecessary allocation. The original &str worked fine;
  push_str(extra.trim()) would too.
• Trailing whitespace in several config.rs doc comments (e.g. /// UVR_CACHE_DIR\n/// ) — rustfmt won't catch these but they're easy
  to clean.

Tests

Could you add a small mod tests in config.rs covering: (1) defaults when env vars are unset, (2) override when set, (3) empty-string
env vars falling through to default? Seven new public entry points with zero coverage is the main thing keeping this from being a
clean merge.

**Out of scope (Bonus!!!) **

uvr doctor should ideally print the effective values of UVR_CACHE_DIR / UVR_R_INSTALL_DIR / UVR_INSTALL_DIR so users can debug "where
did my files go". Happy for that to be a separate PR if you'd rather not expand scope here.

More importantly-> Thank you so much for your help and work here!

[bsirak]

@nbafrank, thank you for the info! I'll make the suggested changes, clean it up, and try to get to adding the output to uvr doctor.

Do you have an opinion on UVR_INSTALL_DIR mismatch handling (error vs warn)? My gut says UVR_INSTALL_DIR should only be used by the installer.sh and the self-update should strictly replace the in current location. If you're OK with it, I want to restore the self_update.rs to before I touched it and allow the installer.sh to continue to use UVR_INSTALL_DIR alone. I know that isn't what we originally discussed, so I'm amenable to anything, but I believe this is current best practice for rust self_update.

[nbafrank]

Go for it regarding UVR_INSTALL_DIR I am new to many of these practices so I welcome your input and feedback!

[bsirak]

hey, sorry. i should've mentioned in my reply that i was just pulling the changes you made into the pr branch. Nothing should've changed yet on my end. I'll post when I think it's ready for another pass.

And no worries, I'm new to most of this too. Certainly on anything this scale.

[nbafrank]

No worries. Please write here when ready! Also there are some failing CI tests so check on those as well. Thanks for the help!

[bsirak]

@nbafrank
ok, I've updated with a first pass on fixes. I'm definitely not done yet, but I think I addressed everything. Couple of things:

1. the testing in env_vars.rs needs a rework. I believe everything will pass right now, but if it ever fails it could go very bad. I'll look at it again tomorrow.
2. I reworked and left UVR_INSTALL_DIR in the env_var.rs crate for now, but it's not used anymore except for the install.sh file. I can remove it if you'd like, but i figured it would be nice to have in the list.
3. I noticed in the cran.rs and p3m.rs crates the cache directory is falling back to the cwd with .unwrap_or_else(|| PathBuf::from(".")). Are we ok with this or should we error when both UVR_CACHE_DIR isn't set and the home directory can't be found?

if you want to take a quick look and let me know if you have any other comments or suggestions, I'd greatly appreciate it. Otherwise, I'll take a second pass at the testing for env_vars.rs tomorrow evening and hopefully remove all unsafe env::set_var calls completely.