[codex] Document route loader transform experiment

[ScriptedAlchemy]

Summary

• Record the route-loader experiment branch as an explicit stack layer above parallel dev compilers.
• Includes the attempted client route transform loader implementation, its revert, and the benchmark note explaining why the loader path was not kept.

Why

This keeps the failed loader investigation visible without mixing that context into the lazy worker-slot or prewarm PRs. The final implementation stays on the custom route transform executor path because the loader experiment was slower in local benchmark runs.

Validation

• Historical branch only; no new local code changes in this restack step.

[github-actions]

Benchmark Results

Compared PR head 7fe442c against base 90ea459.

Total median wall time: 9.93s -> 9.92s (-0.0%, 1.00x speedup)
Compiler ready median: 9.93s -> 9.92s (-0.0%)
Route load median: 2.04s -> 2.03s (-0.2%)

Benchmark	Base total	Head total	Delta	Head ready	Head routes	Speedup	Head RSS p95
large-355-ssr-esm	9.93s	9.92s	-0.0%	9.92s	2.03s	1.00x	-

Profile: large; mode: dev; iterations: 10; warmup: 0.
Workflow run