Fix cross-project bot throttling and event cleanup

[landong-coder]

Bot throttling currently shares an IP/time-window counter across projects and organizations. Its cleanup job also deletes matching events across the entire organization, so one project's traffic can throttle another project or remove events from a project with bot filtering disabled.

Scope counters and batch grouping by organization, project, and client IP. Cache keys use Organization:{organizationId}:Project:{projectId}:bot:{period}:{ip} through a dedicated helper. Cleanup calls RemoveAllByProjectAndClientIpAsync; organization-wide retention cleanup calls RemoveAllByOrganizationAndClientIpAsync. Interface declarations, implementations, callers, and tests use these explicit scope names. Invalid cleanup tasks fail before deletion.

Tests and validation

• Rebased onto upstream main at 7e1c4e9ad.
• Test project builds; eight focused unit tests pass on the latest revision.
• Plugin tests use the existing TestWithLoggingBase, its logger factory, and asynchronous disposal.
• Four repository integration cases cover actual deletion boundaries across projects, organizations, IPs, and optional date ranges. Their names and calls were updated for the explicit repository methods.
• Local integration execution remains blocked at Aspire fixture initialization because Docker is unavailable; deletion assertions have not run locally.
• git diff --check passed.

Compatibility and deployment

Repository interface breaking change: the organization-scoped RemoveAllAsync overload is renamed to RemoveAllByOrganizationAndClientIpAsync, as requested in review. External source callers or custom IEventRepository implementations using that overload must update and rebuild. The project-scoped method introduced in this PR is named RemoveAllByProjectAndClientIpAsync. Existing generic repository deletion APIs remain unchanged. No HTTP API, configuration-key, or work-item payload changes.

The new cache keys start fresh per-project counters; old shared keys expire naturally. Update all event-processing and cleanup worker instances together. Existing queued tasks with valid project IDs remain supported; invalid tasks fail before deletion.

[CLAassistant]

All committers have signed the CLA.

[niemyjski]

Thanks a ton for your pr, these changes would really help a lot and we really appreciate it. I left a few comments which I feel would really help testing and debugging this in production scenarios.