validation: retry a retryable abandon in the continue-as-new backoff test

[Affan Dar (affandar)]

Summary

test_continue_as_new_unregistered_backoff (in src/provider_validation/race_replay.rs, added in #67) failed in CI on the 0.1.32 release PR (#68):

src/provider_validation/race_replay.rs:224: release competing inspection fetch:
ProviderError { operation: "abandon_orchestration_item",
  message: "error returned from database: (code: 6) database is deadlocked", retryable: true }

The test calls the store directly while a runtime fetches from the same store:

Test                                       Runtime (same in-memory SQLite store)
fetch the successor start (to inspect it)
abandon_orchestration_item  ───┐           fetch_orchestration_item
                               └── the two transactions block each other
  → "database is deadlocked", retryable: true
  → test called .expect() → panic          → runtime logged a warning and retried

The validation store is SQLite in-memory with a shared cache (sqlite::memory:?cache=shared). There, two transactions can block each other, and the provider reports a retryable error. The test already retried that error on its own fetch, two lines above, but not on the abandon.

This PR retries the abandon on retryable errors, up to 5 times with a short backoff. instance_locking.rs already does the same for this SQLite error. A failed abandon leaves the lock in place, so the retry is safe.

This code ships in the public provider_validation module. Without the fix, other providers that run the suite could hit the same random failure.

Checklist

• Tests added/updated (fixes a provider validation test)
• Regular tests pass locally (cargo nt), run through ./run-tests.sh
• Doc tests pass locally (cargo test --doc --all-features)
• Docs updated (not needed)
  • Existing docs updated where behavior changed
  • New doc added under docs/ if introducing a new area
  • Linked from docs/README.md

Validation

Rust 1.97.1, macOS.

Run	Result
Before the fix: the test 100 times, 4 loops in parallel	1 failed, same error as CI
With the fix: the test 200 times, 4 loops in parallel	200 passed. The new retry fired 2 times.
This branch: cargo fmt --check, cargo clippy ... -D warnings	pass
This branch: sqlite_provider_validations	232 passed
This fix plus the 0.1.32 release edits: ./run-tests.sh	all features: 1,217 passed. No features: 1,214 passed.
This fix plus the 0.1.32 release edits: cargo test --doc --all-features	46 passed, 74 ignored

No changelog entry: this test is new in the unreleased 0.1.32.

Links / Design notes

• Found on the 0.1.32 release PR: chore(release): duroxide 0.1.32 #68
• Test added in: Fix queue and signal race replay, exact continue-as-new carry, input loss between executions, and KV key order #67
• Separate flaky test found during the same release, tracked for later: worker_reliability_test: crash tests fail at random and never test a crash before the completion is saved #69