v3: extend baseline part2 #5

## Summary This extends the V3 C oracle baseline with the next part of language/runtime coverage needed before continuing native backend work. The PR keeps the work inside the current V3 pipeline and does not add any native backend implementation. ## Changes - Extend V3 parser/checker/transform/markused/CGen support for the part2 baseline cases. - Add regression tests for the newly supported cases around generics, function values, IError, map receivers, string interpolation, numeric literals, filelock helpers, and related CGen paths. - Add the unlocked real examples to `vlib/v3/test_all.vsh` using their repository paths directly. - Include the current V3 C-oracle gate for 35 real cases, including: - `examples/2048/2048.v` - `examples/tetris/tetris.v` - `vlib/v/tests/options/option_test.c.v -autofree` - Add small perf/RSS cleanups: - sparse function-value resolution storage; - scoped cleanup of worker checker caches after parallel transform merge. ## Validation - `git diff --check` - `v fmt -verify` on changed V/VSH files - V3 build with `-d parallel` - Focused regression run: `18 passed, 18 total` ## Performance note `transform (parallel)` is still the main remaining local delta, around `+16%` in my 7-run comparison. RSS/peak memory is back in range. I did not add a broader transform type-resolution cache in this PR, because the remaining hot paths depend on context such as module, scope, smartcasts, aliases, and receiver information. @medvednikov Do you see a better V3-native place to optimize this, or should that be handled in a separate performance-focused PR?