fix(providers): report a reason code when the GJC watcher fails

[devswha]

What changed

• fail() now takes a reason from a fixed vocabulary and attaches it as Error.cause
• child exit reports its numeric exit code and signal name
• reportFailure logs that cause together with the consecutive-failure run length, reset once a start succeeds

before:  GJC native session watcher failed.
after:   GJC native session watcher failed. (child-exit code=1 signal=none; consecutive 47)

Why

A watcher restart loop logs one indistinguishable line per cycle. On a live host it ran 1752 cycles over 14.5 hours and left no record of the cause — the failure was only cleared by an unrelated process kill, and by then the evidence was gone.

Raw stderr is not the fix: watcher diagnostics deliberately never expose transcript paths, an invariant locked by contains callback diagnostics and continues without exposing event paths. A closed reason vocabulary carries no session data and still identifies the failure. Exit code and signal name are numeric/enumerated, so they are safe to attach.

Restart backoff already exists (1s doubling to a 30s cap) and is unchanged; only the run length is new, so a stuck loop is visible without counting timestamps.

Tests

• each failure mode reports its own reason code and nothing from the frame
• ready timeout and child exit stay distinguishable, exit status included
• failure diagnostics name the reason so a restart loop is diagnosable from logs alone
• existing path-leak assertions still pass unchanged
• npm run typecheck, npm run lint, npm run check:identity, npm test (server 907, client 262) all green