Fix timestamp-aware gateway resolution in snapshots and caches

[grandpig]

Summary

Fix gateway resolution so the inserted_at onboarding boundary is consistently enforced by both the snapshot and fallback-cache paths.

Closes #1229

Problem

The per-gateway Trino queries evaluate gateway existence using:

inserted_at <= :timestamp

but the in-memory snapshots do not retain inserted_at, so snapshot hits ignore the supplied query timestamp.

Fallback results are also cached only by public key even though the result depends on both public key and timestamp. This can cause a lookup for one report timestamp to incorrectly determine the result for another timestamp.

Changes

• Include inserted_at in gateway snapshot entries.
• Check inserted_at against the supplied query timestamp on every snapshot lookup.
• Cache timestamp-independent gateway entries instead of final boolean/resolution results.
• Apply the timestamp cutoff after reading from the fallback cache.
• Preserve the existing fail-closed behavior for Trino query failures.

Tests

Added regression coverage for:

• snapshot lookup before onboarding;
• snapshot lookup at or after onboarding;
• fallback lookup before onboarding followed by one after onboarding;
• fallback lookup after onboarding followed by one before onboarding.

These cases are covered for both mobile_packet_verifier and mobile_verifier.

[grandpig]

@michaeldjeffrey Just following up on this in case it got buried. The main semantic change is to retain each gateway's inserted_at in both the snapshot and fallback cache, so resolution remains timestamp-aware regardless of cache order. The regression tests cover both before→after and after→before lookup orderings. Would appreciate a look when you have time. Thanks!