test: observer state-sync frame timeout, monero tests without fsync

Two test-only changes that cut unit test wall time without touching what the tests assert. Each was judged with scripts/rsi/testspeed.py: go test -count=1 -tags mocknet timed 5 times on each side, interleaving base and candidate, with an exact Mann-Whitney test, and every test the base passed had to pass with nothing skipped.

commit	change	measured
test(observer): 250ms frame timeout in the state-sync deadline tests	bifrost/observer is the unit run's critical path. Its state-sync deadline tests wait on multiples of a test frame timeout, and local pipes answer in microseconds, so the timeout drops from 1s to 250ms. The assertions are unchanged.	bifrost/observer 84.5s → 74.3s (−12.1%, p=0.004)
test(monero): write without fsync in tests unless a test counts syncs	The monero stores fsync every write, and a churn rescan was disk-bound: 14s on disk, 0.7s on tmpfs. TestMain switches the stores' write options to plain writes, and the churn tests use the non-syncing scanner storage, which differs only in WriteOptions.Sync on SetScanPos. The one test that counts syncs restores the production options for its run through withProductionFsync. It is not a parallel test, so the swap cannot race, and go test -race is clean. Production still fsyncs.	bifrost/pkg/chainclients/monero 47.3s → 18.4s (−61.2%, p=0.004)

🤖 Generated with Claude Code