Refute global variant population as inference outlier

[gunbai-bot]

Result:

• Measures named candidate . #3 on current main with same-process cold, warm, and padded controls.
• Refutes global_variant_base population: doubling 6,614 to 13,228 entries changes warm target inference by 11ms (0.14%), with structurally equal results.
• Records that current required-floor no longer reproduces the historical subject and does not reach the target.
• Records why the two proposed copied-accumulator edits must not land: PR The tree was lexed four times to compile it once: one acquisition authority, 12,121 tokenizations to 6,061 #9671 migrated the runtime carrier to persistent im::HashMap, making shared clones O(1) structural sharing rather than full-map copies.

Verification:

• Capped local release build under a 24 GiB address-space limit.
• Explicit real importer over all three roots: 720 resolved sources, 3,716 indexed census modules.
• Three target inference arms in one process; cold, warm, and padded outputs structurally equal.
• git diff --check.
• Pre-push cargo fmt --all --check.

All throwaway Rust instrumentation was removed before commit.

[gunbai-bot]

Addressed review 57798 in 8d0a8fe: the plan now names the exact gunbc compile producer invocation, its source-root ordering and entry subject, and the v1.compiler.infer::typecheck_module / infer_items timing seam. It also records the current full-population transformed distribution and scopes the one-off timing implementation honestly. — sent from bright-lynx-473