API improvements for MCP

[nuclearcat]

As discussed in Discord several improvements on Dashboard API is required to provide better efficiency for MCP tools:

1. Wire lab into the existing filter_. mechanism on tree/{commit}/boots|tests|builds, i.e. make filter_boot.lab=lava-collabora / filter_test.lab= / filter_build.lab= actually filter, today they are accepted and silently ignored while filter_architecture and filter_boot.status work, so a lab query has to download all ~34k tests of a commit and discard 99% of them client-side. Accepting the parameter repeatedly (filter_test.lab=a&filter_test.lab=b) for multi-lab selection would match how the UI presents labs.
2. Honour the filter body object on the hardware/{id}/boots|tests|builds|summary POST endpoints, at minimum for lab and status — it is currently ignored in full, so the "which labs run this qcom platform, and how are they doing" question can only be answered by fetching everything for the platform.
3. Add a lab dimension to /api/metrics/: lab_maps gives volumes (builds/boots/tests counts) but no pass/fail, so there is no way to ask "is lab X healthy right now" without going through a specific tree or platform. Per-lab status counts there, ideally with the same PASS/FAIL/ERROR/SKIP shape already used by summary.
   .labs - would make it a one-call answer.