[kvlonge]

Why

Since #1030, a one-model edit recompiles the edited model uncached until native rendering is the default. The release comparison holds every compile command to 1.15x, so its dense one-model edit check failed at 1.232x wall and 1.193x CPU, which would block the next release on an accepted trade-off.

Changes

• New MAX_EDIT_COMPILE_TIME_RATIO of 1.25x, applied to the wall and CPU time of both one-model edit commands (compile (one-model edit) and dense compile (one-model edit)). This mirrors --edit-max-ratio in the same-runner job.
• Cold and unchanged-warm compile commands keep the 1.15x limit; other commands and the RSS limit are unchanged.
• A workflow comment says the edit limit tightens back to 1.15x once native rendering is the default.

Verification

• New verdict tests on the configured commands: edits within 1.25x pass, edits above it fail, and warm and cold compiles still fail above 1.15x
• tests/unit/scripts/release_performance
• make check-ci, hygiene scan