Fix GoldenPath CI and add arithmetic and boolean agg functions

[trevornelson]

Summary

• Fix Zeitwerk autoloader error that was failing CI by moving AUTOLOADER into the Kumi namespace and explicitly ignoring golden_schema_modules.rb
• Add new function definitions: round, floor, ceil, between?, and none? with Ruby and JavaScript kernels
• Add kernel finalize support to allow reducers whose final result differs from their accumulator (used by none? which accumulates with OR then inverts)
• Add fold_inline to boolean aggregate kernels (any, all, none) to fix "Can't fold - no template defined" error when these reducers are applied to collection types

Autoloader fix

The top-level AUTOLOADER constant caused Zeitwerk namespace errors in CI. I moved it inside the Kumi namespace and added an explicit ignore for golden_schema_modules.rb, which is now loaded via require_relative only when golden tests are invoked.

New functions

Function	Kind	Description
round(value, precision)	elementwise	Round to N decimal places
floor(value)	elementwise	Round down to integer
ceil(value)	elementwise	Round up to integer
log(value)	elementwise	Calculate the normal log
between?(value, lo, hi)	elementwise	Inclusive range check, returns boolean
none?(source_value)	reduce	True when no element is truthy

Kernel finalize step

The none? reducer accumulates the same way as any? (boolean OR), then inverts the result. To support this, a finalize attribute was added to the kernel schema. During codegen, emit_loadaccumulator in both Ruby and JS emitters now checks for a finalize template and applies it to the accumulator before assigning the output.

Boolean aggregate fold_inline

When a reduce function is called on a collection type (tuple), the compiler generates a Fold LIR instruction. emit_fold requires a fold_inline template on the kernel. The numeric aggregates (sum, min, max, count) already had this, but the boolean aggregates did not, causing a runtime error. Added fold_inline for all three boolean aggregates.

[trevornelson]

finalize was needed so that none? could be calculated correctly when the function is used as an accumulator.