fix(mcp): get_anomaly keeps the events nearest the occurrence and says how many it left out (#5861)

Each source in an anomaly's `around` was cut to its oldest 40 with
nothing in the reply to say so; on a busy host the events nearest the
occurrence were the ones dropped. Each source now keeps the 40 nearest
the occurrence's start, and `around_left_out` counts the rest. Sources
return everything their bounded query read (Docker and the host agent
no longer cut at 40 and 20 themselves), so the count is honest.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

docs/plugins/writing-a-plugin.md

@@ -403,7 +403,7 @@ async def mcp_events(db, host_id, host_name, since, until) -> list[dict]:
             for r in await _jobs_on(db, host_id, since, until)]
 ```
 
-- Return what happened between `since` and `until`, oldest first: something that lasted gives `until`. Keep it to tens of rows; the caller keeps the first 40.
+- Return what happened between `since` and `until`, oldest first: something that lasted gives `until`. Bound your query (a few hundred rows); Steward keeps the 40 nearest the occurrence's start and tells the caller how many it left out (`around_left_out`).
 - It is called only while the plugin is enabled, with its own session. If it raises, the anomaly is still returned without the plugin's events. The host agent (maintenance jobs, reboots) and Docker (container starts, deaths, OOM kills) are the examples to read.
 
 A resource with no host (a Traefik router, say) has no host to read events by. A plugin that knows what happened to one of its own resources exports `mcp_resource_events(db, source, resource, since, until)` instead, same return shape, and returns `[]` for a source that isn't its own. Traefik's is the example: it gives a router's swarm service's container events, on whichever hosts they ran.