fix(client): coalesce catalog refetches after event bursts

[Hona]

A 33-minute Desktop netlog showed GET /api/model fetched 16 times (~270 KB each, 4.3 MB decoded) and GET /api/provider 16 times. One Location's model and provider catalogs were fetched three times within ~190 ms because catalog.updated, integration.updated, and credential.updated each answered with an immediate refetch. /api/command (24), /api/mcp/resource (22), /api/agent, /api/skill, and /api/reference saw the same burst pattern.

sequenceDiagram
  participant S as Server
  participant D as createData
  participant A as /api/model
  S->>D: catalog.updated
  S->>D: integration.updated
  S->>D: credential.updated
  Note over D: invalidate now, one settle timer per resource + Location
  D-->>D: 150 ms quiet
  D->>A: GET /api/model (once)

Loading

Changes — packages/client/src/solid/data.ts

• Every per-Location catalog event now goes through settle(key, load) instead of refresh(load): catalog.updated, integration.updated, credential.updated, credential.switched, agent.updated, command.updated, skill.updated, reference.updated, config.updated, websearch.updated, and mcp.resources.changed.
• Settle keys mirror the sync keys, location.<field>:<locationKey>, so different events that ask for the same catalog share one timer. catalog.updated and integration.updated for the same Location now coalesce into one model fetch and one provider fetch.
• invalidate(...) still runs immediately, so an explicit sync() during the window reads fresh data and already-published data stays visible until the reload lands. Session, permission, form, shell, and VCS handling is unchanged.

case "integration.updated":
  result.location.integration.invalidate(location)
  result.location.model.invalidate(location)
  result.location.provider.invalidate(location)
  settle(`location.integration:${locationKey(location)}`, () => result.location.integration.sync(location))
  settle(`location.model:${locationKey(location)}`, () => result.location.model.sync(location))
  settle(`location.provider:${locationKey(location)}`, () => result.location.provider.sync(location))
  break

Scenario	Before	After
catalog.updated + integration.updated + credential.updated for one Location in ~190 ms	3 × GET /api/model, 3 × GET /api/provider	1 × GET /api/model, 1 × GET /api/provider
33-minute Desktop session	16 × GET /api/model (4.3 MB decoded), 16 × GET /api/provider	1 fetch per burst per Location
/api/command, /api/mcp/resource, /api/agent, /api/skill, /api/reference	one fetch per event (24 and 22 for the first two)	one fetch per burst per Location

Follows #47561, which introduced settle for mcp.status.changed.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.