codegen, text: remove vestigial Option/Result wraps and redundant clones

[iainmcgin]

Follow-up to #121: addresses the remaining actionable findings from the code-style audit — clippy::unnecessary_wraps (4 sites) and clippy::redundant_clone (2 sites), both nursery lints.

Changes

buffa-codegen/src/message.rs::singular_skip_predicate — returned Option<&'static str> but every match arm was Some(...). Now returns &'static str; the caller skip_serializing_predicate wraps it for the one branch where the value is conditional on field shape (the other branches already produce their own None/Some). A doc comment records that singular fields always have a default-value predicate.

buffa-codegen/src/view.rs::repeated_to_owned — returned Result<TokenStream, CodeGenError> but the body was a single infallible match. The signature implied a failure mode it didn't have, and the caller's ? was propagating an error that could never occur. Now returns TokenStream directly, which makes the asymmetry between it and the genuinely-fallible sibling singular_to_owned honest.

buffa/src/text/token.rs::emit / ::emit_eof — returned Result<Token<'a>, ParseError> purely so the FSM's return self.emit(...) tail calls didn't need an Ok(...) wrap. The 12 call sites now wrap explicitly, which makes the success/failure partition of parse_next visible at every return: Ok(self.emit(...)) always succeeds, Err(self.err_here(...)) always fails, self.parse_*() delegates. Before, the success path's Result typing was implicit while the error path's was explicit.

buffa-codegen/src/message.rs::compute_field_info — two bytes_type.clone() calls in mutually-exclusive if/else if branches alongside a third branch that already moved the value. Both clones removed; each execution path moves bytes_type at most once.

Notes

No change to generated output — task gen-wkt-types produced no drift. task lint and task test both pass. Net 47/-45.

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}