Add Executor.stream() async generator and ModuleAnnotations.streaming

[tercel]

Summary

Add optional stream() async generator to Executor and streaming: bool to ModuleAnnotations, enabling modules to produce incremental output.

Changes

ModuleAnnotations

• Added streaming: bool = False to frozen dataclass

Executor.stream()

• New async def stream(module_id, inputs?, context?) -> AsyncIterator[dict]
• Steps 1-6 identical to call_async() (context, safety, lookup, ACL, validate, before-middleware)
• Inline async-aware middleware dispatch (matches call_async() pattern with inspect.iscoroutinefunction)
• If module has no stream(): falls back to _execute_async(), yields single chunk
• If module has stream(): iterates it, yields each chunk, accumulates via {**accumulated, **chunk}
• Output validation + after-middleware on accumulated result
• Error recovery via _execute_on_error_async()

Tests (5 new)

• Fallback single chunk, multiple chunks, ModuleNotFoundError
• Before/after middleware ordering
• Disjoint-key shallow merge accumulation

Files Changed

• src/apcore/module.py — streaming field
• src/apcore/executor.py — stream() method
• tests/test_executor_stream.py — 5 new tests

Test Results

957 tests pass, 0 failures.