fix(alerts): sync cloud read-state to desktop reliably

[proofofprints]

Summary

• Acking an alert in the cloud portal/mobile app wasn't marking it read on desktop, because the cloud's alert-read / alerts-read-all WS payloads couldn't be matched against local AlertEvents.
• Two compounding bugs, fixed together with the OverCloud-side fix (overbuildlabs/OverCloud, same branch):
  1. minerId always null: the desktop posts alerts with the miner identifier under minerIp, but OverCloud's ingest schema only accepted minerId — so every alert from desktop landed in Postgres with minerId: null, and the cloud could never echo back a matching identifier.
  2. timestamp precision loss: desktop alerts carry nanosecond-precision RFC-3339 timestamps; round-tripping through Postgres (truncates to microseconds) and the WS broadcast re-encoding (milliseconds) drops precision, so exact-instant comparison in timestamps_match never matched.
• This PR fixes Add device removal to ASIC and Mobile miners #2: timestamps_match now compares at whole-second granularity (DateTime::timestamp()), which is safe given alert cooldowns are 15–30 minutes — no realistic case where the same rule+miner fires twice in the same second.

Test plan

• Existing unit tests in src-tauri/src/commands/alerts.rs (matches_on_rule_miner_and_timestamp, rejects_on_any_field_mismatch, timestamp_match_is_format_tolerant, mobile_device_id_matches_as_miner_id) cover the matching logic and remain valid against the new comparison granularity.
• Manual: ack an alert in the cloud portal/mobile app and confirm it shows read on desktop within one sync cycle, and vice versa.
• Note: full cargo test/build isn't runnable in this sandbox (missing system GTK/GDK dev libs for the Tauri GUI deps) — verify in CI/locally.

Companion fix: overbuildlabs/OverCloud#35 (commit 6bd0ddd, already on that PR's branch).

---

Generated by Claude Code