grafana: filtered events per node

[s0me0ne-unkn0wn]

Closes #31.

The TART Global dashboard lets you pick an event group / event type, but there was no per-node breakdown of the selected events — no way to see which nodes are the top senders.

Backend

New endpoint GET /api/grafana/events-by-node: per-node totals for a set of event types over a range, ranked by count, each row joined with the node's handshake identity (address, implementation name/version, connection state, last seen) and its share of the network-wide total (computed over all nodes before limit, so shares stay meaningful when the list is truncated).

• event_types accepts the same mix as /timeseries (numeric IDs, event names, group names, Grafana {a,b} braces); omitted = all types.
• interval is a resolution hint that picks the aggregate tier (30 s counts / 1 m / 1 h aggregates); omitted, it is derived from the range length (~300 buckets, mirroring Grafana's $__interval).
• Aggregation happens in SQL (GROUP BY node_id over the same tiers /timeseries reads), so the panel gets ≤ limit rows instead of buckets × nodes.

Tier selection is factored into a shared select_event_stats_table() used by both /timeseries and the new endpoint. Side fix: sub-minute intervals on ranges older than 30 days previously selected the _1m aggregates whose retention had already expired; they now fall back to _1h.

Grafana (TART Global, "Selected Events" row)

• Selected Events by Node — timeseries of the selected event types with group_by=node_id, legend table sorted by total, 64-hex node ids shortened.
• Top Senders (Selected Events) — table ranked by count: Node ID (links to the TART Node dashboard), Events (gauge), Share, Address, Impl, Version, Connected, Last Seen.

🤖 Generated with Claude Code (but I read it through and cut some slop, honest!)

[michalkucharczyk]

I am still not convinced if we need new endpoint, yeah, but let's keep it.
Maybe it shall be named events/by-node to match naming convetion?