Implement the 0.38 performance tranche

[rbroderi]

Summary

• add frame-free pure recursive base cases and scalar compiled-call entries
• prove guarded dense primitive-write regions
• reduce fresh record construction and lazy table metadata costs
• add the 0.38 benchmark, correctness coverage, documentation, and version bump

Performance

• balanced recursion: 22.0% faster on Python 3.13 and 18.7% faster on Python 3.14
• recursive record continuations: 5.4% and 3.2% faster
• dense aliased primitive writes: 1.6% and 4.1% faster
• controls remained within the 5% regression gate

Validation

• 598 Python tests on CPython 3.13 and 3.14
• official Lua 5.5.1 probes: 24/24 on both versions
• Penlight 23/23, luatest 5/5, LuaCov scanner 24/24, Are We Fast Yet 11/11 on both versions