android: decode graph patches off the main thread during restore

[tiensonqin]

Summary

Cold-launch profiling on the 1000-journal/100-block seeded graph showed backend.applyJson (JSON parse + LuiRetainedTree mutation) running on the main thread for 464ms on a 461KB graph patch — the largest single contributor to the "Skipped 75+ frames" stall between authentication_published and journals_ui_ready.

applyCoreResponse now runs bridge.applySnapshot + applyPatch inside withContext(patchDispatcher) where patchDispatcher = Dispatchers.Default.limitedParallelism(1): decode + tree mutation happen off the main thread while patch ordering stays serial (the dispatcher is single-threaded, so concurrent applyCoreResponse / onPatch calls still apply in order). Publishing stays safe — LuiBackend.publish only writes Compose snapshot state, which is designed for cross-thread writes.

Also adds a GraphLifecycle timing log (applyJson took Xms for N chars, only when >20ms) — this is what attributed the stall and is useful for future launch profiling.

Measured on the emulator (1000×100 graph): the 461KB patch moved from 464ms on main to 289ms on a background worker; total journals_ui_ready is bounded by the remaining Compose frame, which this also unblocks.

Link to Devin session: https://app.devin.ai/sessions/0f103e5f2f014fa887d43048a7c1200f
Open in Devin Desktop: https://app.devin.ai/desktop/session/0f103e5f2f014fa887d43048a7c1200f?variant=devin
Requested by: @tiensonqin

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring