fix(client): await_ready retries on transient errors instead of bailing

[githubrobbi]

Summary

Aligns the sync UffsClientSync::await_ready with its async sibling: any non-Ready, non-success outcome from a status poll (Loading status, I/O error, connection closed, RPC timeout, transient protocol error) now keeps polling until the caller-supplied timeout deadline elapses.

Root cause

crates/uffs-client/src/connect_sync.rs::await_ready matched Err(other) => return Err(other), so a single transient error from any in-flight status RPC aborted readiness probing immediately. The async sibling at crates/uffs-client/src/connect.rs::await_ready has always classified non-Ready, non-success outcomes as PollOutcome::OtherError and continued the loop — the sync path was the outlier.

Why now — 2026-05-07 Phase 7 soak finding

The Phase 7 24-h soak attempt failed with:

Error: Daemon did not become ready in time
  Caused by: request timed out

even though the captured daemon.log showed the daemon up and IPC-listening 1.3 s after spawn. The very first status RPC during the Windows AF_UNIX socket-bind window can hit its per-RPC deadline and surface as ClientError::Timeout, which the pre-fix code returned immediately — aborting probing while the daemon was healthy and one poll away from Ready.

PR #146 already added a soak-harness-level workaround (idempotent attach + race-tolerant spawn). This PR fixes the underlying CLI bug so direct uffs daemon start invocations no longer race.

Regression test

await_ready_retries_on_protocol_error_until_deadline (crates/uffs-client/src/connect_sync_tests.rs) feeds a JSON-RPC error response to the very first poll, surfaced as ClientError::Protocol.

• Pre-fix: test fails in <1 ms with Err(Protocol("transient mid-handshake error"))
• Post-fix: test takes the full 120 ms deadline and returns Err(ClientError::Timeout)

The 0.332 s observed test wall-clock confirms the loop actually polled.

Local validation

• ✅ cargo nextest run -p uffs-client — 173/173 passed, including the new test and all four existing await_ready_* regression pins
• ✅ just lint-fast — fmt-check, file-size, typos, reuse, lint-ci, lint-prod, lint-tests
• ✅ just check-windows — cargo xwin check --workspace --all-targets --all-features

Compliance audit (mandatory rules)

1. No suppression hacks — no allow / cfg / ignore added; the fix is one match arm. File-size policy on connect_sync.rs was avoided by trimming an inline comment (the regression test's docstring carries the rationale), not by adding a file_size_exceptions.txt entry.
2. Surgical, correct fixes — single match arm changed; mirrors the async sibling's existing PollOutcome::OtherError classification.
3. Preserve behavior & contracts — await_ready still returns Err(ClientError::Timeout) on deadline expiry; only the fast-fail-on-transient-error edge case changes (and that path was the bug).
4. Improve tests, don't dodge them — new regression test pins the contract; all existing await_ready_* tests still pass.