Add HTTP/2 PING keepalive to prevent idle connection timeouts

[balaji-g]

Summary

Many cloud load balancers and network proxies terminate idle TCP connections after a fixed timeout. HTTP/2 PING frames (RFC 9113 §6.7) are the standard mechanism to keep connections alive, but httpcore does not currently send them — meaning any HTTP/2 request that takes longer than the idle timeout will fail with a disconnect error.

This adds a background keepalive thread that sends periodic HTTP/2 PING frames on active connections. It can be enabled via:

• Constructor parameter: h2_ping_interval on ConnectionPool / HTTPConnection / HTTP2Connection
• Environment variable: HTTPCORE_H2_PING_INTERVAL (value in seconds)

The env var approach is particularly important because it allows users of higher-level clients (httpx, OpenAI Python SDK) to enable PING keepalive without any code changes.

Example

import httpcore

# Via constructor
pool = httpcore.ConnectionPool(http2=True, h2_ping_interval=60)

# Or via environment variable (no code changes needed)
# export HTTPCORE_H2_PING_INTERVAL=60
pool = httpcore.ConnectionPool(http2=True)

Design decisions

• Daemon thread: The PING loop runs in a daemon thread using threading.Event.wait() for the interval, avoiding interference with the async event loop.
• Thread safety: PING writes are protected by a dedicated threading.Lock to avoid races with the main h2 state machine.
• Env var fallback: HTTPCORE_H2_PING_INTERVAL is read at connection init time. An explicit constructor value always takes precedence.
• Graceful shutdown: The PING thread is stopped and joined on close() / aclose().

Related issue: #1080

Checklist

• I understand that this PR may be closed in case there was no previous discussion. (Discussion in Add optional HTTP/2 PING keepalive to prevent idle connection timeouts #1080)
• I've added a test for each change that was introduced, and I tried as much as possible to make a single atomic change.
• I've updated the documentation accordingly.