[ZeroPointSynergy]

Fixes #1433 (SQLite backend part).

Summary

The in-memory gossip backend already caps distinct relays per public key at MAX_RELAYS_PER_PK (+1, mirroring ). The SQLite backend had no such cap in , allowing unbounded growth of rows via a connected relay.

This change mirrors the in-memory cap in the SQLite backend: skip inserting a new relay once the per-key count exceeds MAX_RELAYS_PER_PK, and propagate DB errors instead of masking them.

Verification

• cargo test -p nostr-gossip-sqlite: 9/9 pass
• adversarial review applied (off-by-one + error masking fixed)

[codecov]

Codecov Report

❌ Patch coverage is 97.64706% with 2 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
gossip/nostr-gossip-sqlite/src/store.rs	97.64%	2 Missing ⚠️

📢 Thoughts on this report? Let us know!

[ZeroPointSynergy]

Added a test to cover the relay cap, since the codecov patch check was failing (81.25% vs 81.76% target — the new SELECT COUNT(*) and the early return Ok(()) were uncovered).

test_relay_cap_per_public_key asserts, against the SQLite backend:

• a public key stores at most MAX_RELAYS_PER_PK + 1 distinct relays (mirroring the in-memory backend, which accepts while relays_len <= MAX_RELAYS_PER_PK);
• further distinct relays are skipped rather than growing relays_per_user;
• already-known relays keep being updated once the key is capped (a second event from the same relay increments received_events), so the cap does not suppress updates;
• the cap is scoped per public key, so a capped key cannot starve another key's relay list.

I mutation-tested the test rather than trusting a green run — each of these broken implementations makes it fail:

Mutation	Result
cap condition removed	11 rows vs 8 expected
> changed to >=	7 rows vs 8 expected
if !exists wrapper dropped	received_events 1 vs 2 expected
COUNT query not scoped to the key	second key 0 rows vs 8 expected

cargo test -p nostr-gossip-sqlite: 10/10 pass. cargo fmt --check and cargo clippy --all-targets -- -D warnings are clean on the pinned 1.85.0 toolchain. I also let cargo fmt fix a formatting drift in the exists query introduced by the previous commit.