PyShen S42 fast build 0.3.0

This is a separate optimized build. Keep the previous 0.2.1 build as the semantic reference.

Optimizations:
- remove exception-driven misses from lexical environment lookup
- use interned Shen symbols with Python object identity hashing/equality
- direct Shen-to-Shen closure jumps inside the evaluator loop, avoiding Tail allocation on every call
- retain the 10,000 Python recursion limit required by the full test corpus

Correctness checks performed locally:
- pytest regression suite: 17/17 passed
- S42 boot and REPL smoke tests
- prime, YACC/type annotation and Montague cases previously responsible for failures

Representative benchmark in this environment:
  (prime*? 100003)
  reference interpreter: roughly 6-7 seconds
  fast interpreter:      roughly 4 seconds

The reference build remains valuable: use it to resolve any discrepancy introduced by optimization.
