Add ASGI marshalling optimizations for reduced Erlang-Python overhead

Implement optimized ASGI request/response handling:

- Pre-interned Python string keys for ASGI scope dicts (15+ keys)
- Cached constant values (http type, HTTP versions, methods, schemes)
- Thread-local response pooling (16 slots per thread, 4KB initial buffer)
- Direct ASGI NIF path bypassing generic py:call()
- Zero-copy body handling threshold (1KB)

New files:
- c_src/py_asgi.h: Header with interned key declarations and types
- c_src/py_asgi.c: Implementation of all optimizations
- src/py_asgi.erl: High-level Erlang API

Expected performance gains:
- Interned keys: +15-20% throughput
- Response pooling: +20-25% throughput
- Direct NIF: +25-30% throughput

Author: benoitc