Remove global read_timeout default, add watcher-level idle timeout

[doxxx93]

Fixes #1798

Motivation

The default config.read_timeout (295s) is applied at the hyper-timeout connector level, which enforces an idle timeout on all TCP I/O indiscriminately. This breaks long-lived connections like exec, attach, and port-forward — if there's no stdin/stdout activity for 295s, the connection is killed with a broken pipe.

The Go client has no global read timeout. Watch streams rely on the server-side timeoutSeconds parameter instead.

Verified on a kind cluster (v1.35.0): exec connections die after idle with read_timeout set, but survive indefinitely with None.

Solution

1. Default read_timeout to None in all Config constructors, matching the Go client behavior.

2. Add a watcher-level idle timeout (next_with_idle_timeout) that wraps stream.next() with tokio::time::timeout. The timeout is set to the server-side timeoutSeconds + 5s margin, so watches still recover from dead connections where the server's close never arrives.

This way each timeout is owned by the layer that actually needs it:

• exec/attach/portforward: no timeout (works indefinitely)
• watch streams: watcher manages its own idle timeout (reconnects on network failure)

[codecov]

Codecov Report

❌ Patch coverage is 73.91304% with 6 lines in your changes missing coverage. Please review.
✅ Project coverage is 76.5%. Comparing base (cfa38f2) to head (872c203).
⚠️ Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
kube-runtime/src/watcher.rs	74.0%	6 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##            main   #1945     +/-   ##
=======================================
- Coverage   76.5%   76.5%   -0.0%     
=======================================
  Files         89      89             
  Lines       8566    8587     +21     
=======================================
+ Hits        6547    6563     +16     
- Misses      2019    2024      +5     

Files with missing lines	Coverage Δ
kube-client/src/config/mod.rs	54.7% <ø> (ø)
kube-runtime/src/watcher.rs	58.3% <74.0%> (+1.9%)	⬆️

... and 6 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[doxxx93]

next_with_idle_timeout had no tests, so patch coverage was at 33%. Added 3 tests using #[tokio::test(start_paused = true)] — tokio auto-advances virtual time when the runtime is idle, so the 295s timeout completes instantly.

• returns_item_when_stream_has_data — data available, returns immediately
• returns_none_on_dead_connection — stream hangs forever, timeout fires → None (triggers reconnect)
• returns_none_when_stream_ends — stream closes normally → None

[clux]

Thanks for this. Sorry took me a while to get to it.
Left a few comments on some documentation choices, but otherwise the wrapping function here is a very nice solution.