GeezerVison

What this MR fixes

This MR resolves async networking regressions introduced while fixing large-response behavior, and stabilizes HTTP/SSL mocking across httpx and aiohttp.

1) Large async response hang (test_async_case_high_number)

• Fixed a deadlock caused by writing large mocked responses directly into OS pipes.
• Introduced a hybrid pipe strategy:
  • small responses mirror bytes into pipes (legacy-compatible behavior),
  • large responses use bounded readiness signaling to avoid pipe-buffer blocking.
• Kept response payloads in shared in-memory buffers per address.

2) aiohttp SSL failures (test_aiohttp)

• Fixed ServerDisconnectedError / SSL read-loop instability by decoupling SSL reads from shared socket cursor state.
• MocketSSLSocket now keeps a private pending SSL read buffer snapshot after writes, so protocol reads are deterministic across runtimes.
• Preserved expected pre-write SSL probing behavior (SSLWantReadError before first write after handshake).

3) Multi-socket regression during full test run

• Fixed TypeError: 'NoneType' object cannot be interpreted as an integer in recv() when called via recv_into() with buffersize=None and pipe-backed reads.
• recv() now normalizes None to the socket default buffer size.

4) PyPy compatibility regression (BufferError)

• Fixed BufferError: Existing exports of data: object cannot be re-sized seen in true-send HTTP flows on PyPy.
• Root cause: BytesIO.getbuffer() exports remained active while later writes resized the buffer.
• Replaced resize-sensitive getbuffer() reads with getvalue()-based logic in socket buffering paths.

5) Coverage recovery for new logic

• Added targeted tests for:
  • shared per-address I/O behavior,
  • pipe-mode flags and reset behavior,
  • large-response readiness sentinel drain/resync path,
  • SSL read behavior around handshake/write ordering,
  • empty server_hostname handling in SSL bio wrapping,
  • getpeercert() fallback to Mocket._address when host/port are unset.

Result

• Local full test target passes (make test).
• Previously failing async paths (httpx high payload + aiohttp SSL flow) now pass consistently.
• PyPy-specific true-send path is stable again.
• Coverage remains high (overall 99%; mocket/socket.py and mocket/ssl/socket.py at 100%).
• Latest local run: 228 passed, 3 xfailed (+ tests/test_pook.py passed).