test(mock): observe fetch RPC timeout firing on a silent later chunk (TEST-7)

[fornwall]

TEST-7 — Fetch timeout never observed firing

Adds fetch_timeout_fires_on_a_reader_hung_on_a_silent_stream to tests/mock_spanner.rs, pinning that a hung later-chunk fetch surfaces Status::Timeout when spanner.rpc.timeout_seconds.fetch is set — the deadline spawn_prefetch wraps each pull_chunk in, previously unexercised end-to-end through the wire.

What the test does

• Reuses the silent-stream script from cancel_unblocks_a_reader_hung_on_a_silent_stream, now factored into a shared silent_after_first_row_server() helper (one row, then the stream's sender is parked alive so it never ends). The cancel test's own doc already called this "the foundation for future timeout tests".
• Sets spanner.rpc.timeout_seconds.fetch=0.5 with one row per batch: the first next() yields the buffered row; the second — the later-chunk fetch that blocks forever — must return an ADBC error with Status::Timeout naming the fetch option.
• A >= 0.5s floor (timed from before execute, since the prefetch's second fetch begins during execute) rules out an unrelated transport failure racing the deadline.

This is the gating twin of resilience's Toxiproxy timeout assertion (its cancel twin already existed as the mock silent-stream cancel test): it needs no proxy, so it runs offline in every cargo test.

Non-vacuity

Verified by disabling the deadline (fetch=0.0): the second next() then hangs forever (the cancel twin proves the same silent stream never ends on its own) and the Watchdog fires — only the driver's deadline can end it.

Checks

• cargo fmt --all --check — pass
• cargo clippy --all-targets --all-features -- -D warnings — pass
• cargo test --test mock_spanner — 39 passed (new test in 0.51s; refactored cancel test still green)

🤖 Generated with Claude Code

https://claude.ai/code/session_01EaKycJ7J9wtPMYcvqSGmck