fix(async): apply socket_timeout per read in async parsers

[goingforstudying-ctrl]

Fixes #3454

The async client was wrapping the entire parser.read_response() call in async_timeout, so socket_timeout applied to the whole response. The sync client applies it per socket recv() instead.

This change passes the timeout down to the async parsers (RESP2, RESP3, and Hiredis) and applies it around each individual stream read, matching the sync semantics.

---

Note

Medium Risk
Changes core async I/O and timeout behavior for all connections (RESP2/RESP3/Hiredis), which can alter when slow large replies time out versus succeeding; buffer handling for pipelining was also adjusted.

Overview
Fixes #3454 by aligning async socket_timeout with the sync client: the timeout applies to each stream read while parsing a reply, not to the whole read_response() call.

Connection.read_response no longer wraps the parser in a single outer async_timeout; it passes read_timeout into parser.read_response(timeout=...).

Async parsers (Python RESP2/RESP3 base, Hiredis) accept a timeout argument and apply async_timeout around individual stream.read() calls via new _read_from_stream. RESP2/RESP3 recursive parsing threads timeout through _readline / _read. Hiredis applies it in read_from_socket and on recursive push handling.

The Python async RESP buffer _clear now keeps unconsumed bytes after _pos for pipelining, with a full buffer reset on on_connect. _read / _readline were reworked to fill the buffer incrementally instead of relying on readexactly for whole segments.

New tests in test_async_timeout_per_read.py cover multi-chunk success, per-chunk failure, nested arrays, SENTINEL default, Hiredis, and connection-to-parser timeout forwarding.

^{Reviewed by Cursor Bugbot for commit 54d8a65. Bugbot is set up for automated code reviews on this repo. Configure here.}

[petyaslavova]

Hi @goingforstudying-ctrl, thank you for your contribution! I'll review it shortly.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Reviewed by Cursor Bugbot for commit d159301. Configure here.}

[goingforstudying-ctrl]

Hi @petyaslavova, this PR (async socket_timeout per-read fix) has been waiting for CI approval. Would you be able to approve the workflow run when you get a chance? Thanks!

[goingforstudying-ctrl]

Addressed the review comments:

1. Pipelined data discarded after parse — fixed by preserving bytes beyond _pos in _clear()
2. Reconnect retains stale parser buffer — fixed by doing a full buffer reset in on_connect() instead of calling _clear()

Both issues have been resolved in the latest commits.

[goingforstudying-ctrl]

Hi @petyaslavova, I addressed the review issues in the latest commits and CI looks green now. Would you have a chance to take another look? Thanks!