fix: recover ldk restart after background timeout

[ovitrif]

What happened?

Backgrounding Bitkit about one second after tapping Restart can make the restart's start() call stall in Electrum fee-rate estimation until its approximately 15-second deadline. The timeout leaves the node lifecycle in ErrorStarting, shows a failure toast, and requires the next foreground resume to trigger recovery.

Unrelated Ktor requests continue succeeding during the stall. The next fee-rate cache update completes in approximately 680 ms.

Expected behavior

The restart flow should recover automatically from a transient fee-rate estimation timeout and return the node to Running without another foreground cycle or manual action.

Steps to Reproduce

1. Open Settings → Advanced → Dev Settings → LDK.
2. Tap Restart.
3. Background Bitkit approximately one second later.
4. Wait for the restart attempt to finish.
5. Foreground Bitkit and inspect the node state.

Logs / Screenshots / Recordings

Starting up LDK Node …
Updating fee rate estimates timed out: deadline has elapsed
Failed to start node during restart
LDK Node error: Feerate estimation update timeout.
Fee rate cache update finished in 682ms.

Joao's review evidence includes the full run details and the known-failing restart-while-backgrounded.xml journey.

Bitkit Version

2.4.0 development build from #1122.

Device / OS

Pixel-class arm64 emulator, Android API 36.

Reproducibility

Always — 3/3 with com.synonym:ldk-node-android:0.7.0-rc.62 and 2/2 with 0.7.0-rc.58.

Additional context

restartNode() calls start(shouldRetry = false). A transient start failure therefore commits ErrorStarting immediately, while the next foreground-triggered start can complete the same fee-rate work successfully.

Acceptance criteria

• Backgrounding Bitkit during Restart does not leave the node in ErrorStarting.
• A transient fee-rate estimation timeout triggers bounded automatic recovery.
• Successful recovery returns the node to Running without another foreground cycle or manual restart.
• Recovered transient failures do not show a terminal restart-failure toast.
• Automated coverage reproduces the background-during-restart sequence and verifies node usability after recovery.