Retry Linux connections promptly after network recovery or resume

[vbmithr]

After suspend or a local network outage, chatd could remain offline for minutes even after networking returned: shared reconnect backoff grew to 15 minutes and no recovery event interrupted the wait.

On Linux, wake that wait on GIO network-availability/configuration hints or login1 resume signals, reset accumulated backoff, and retry promptly. GIO supports netlink without NetworkManager. A single GLib event thread serves all accounts; the added Linux-only Rust dependency already exists in the lockfile.

Recovery hints remain hints: they do not mark accounts Connected, close healthy connections, revive expired sessions, or bypass ordinary HTTP/command pacing. Attempts stay at least one second apart, including event bursts, and continued failures still back off. Other platforms retain timed reconnection; remote DNS/service recovery with no local OS event is not detected by this change.

Includes docs/reconnecting.md with behavior, limitations, restart instructions, and repeatable tests.

Validation:

• Daemon unit suite with member changes: 137 passed, 1 manual namespace test ignored.
• Independent clean checkout of the original fix: 134 passed, 1 ignored.
• Real kernel route notification through GIO's netlink backend passed in a disposable network namespace, with host networking untouched and no NetworkManager required.
• Synthetic real-daemon integration suites: Discord 31 passed (4 manual audit tests ignored), Signal 10, Slack 15, Teams 7, Messages 5.
• Arch package revision 13 built successfully.
• Rechecked daemon unit tests after applying the commit to current main.

Fixes #41. The Discord viewport-reconciliation findings in #42/#43 and subscription-budget finding in #44 are separate.

[itz4blitz]

Linux reconnect waits now wake on GIO network-available and login1 resume, reset backoff, and still keep a one-second floor so hint bursts cannot storm. Hints do not mark an account connected, close a healthy socket, or revive an expired session. The Windows failure was a_call_that_moves_to_another_voice_server_keeps_its_devices_and_goes_on seeing 5 packets instead of 15; that test is in mediad and this branch does not touch it. A rerun of the failed job passed.