Emit rum.session_id as multi-value tags, capped at 10

[r1viollet]

Summary

• Adds rum.session_id:<uuid> profile-level tags, one entry per completed session record produced during the profile window. Repeated entries with the same key (the standard Datadog multi-value facet shape — not comma-joined, not a JSON list) so each session id is independently queryable as @rum.session_id:<uuid>.
• Placed alongside rum.application_id in PrepareAdditionalTags. Reuses the existing AddSingleTag push helper. Both rum.* tags now live in the same code region (ProfileExporter.cpp:1592-1620).
• Cardinality capped at MAX_RUM_SESSION_ID_TAGS = 10 to bound tag indexing cost on pathological session-rotation patterns. Overflow logs Log::Warn reporting how many ids were skipped; the full id list still travels in optional_internal_metadata_json (rum_session_ids), unaffected by the cap.
• Records are 1:1 with sessions by construction (Profiler::CompleteCurrentSession pushes once per session and clears _currentSessionId), so no dedup is needed in the collection step.

Implementation notes

Tag-list construction is factored into a pure static helper ProfileExporter::BuildRumTags(applicationId, sessionIds) -> vector<pair<string,string>>. ddog_Tag is opaque, so we can't inspect entries back out of ddog_Vec_Tag after pushing — pinning the contract upstream of the push is the only way to write meaningful unit tests.

Export() mirrors the session ids that already feed optional_internal_metadata_json into a separate rumSessionIds vector before clearing _sessionRecordsBuffer, then threads that vector down through ExportProfile to PrepareAdditionalTags.

Out of scope

The customData["view.id"] array (matching the chrome SDK contract for the runtime view-filter endpoint) and the optional_info_json channel are intentionally not in this PR.

Test plan

Unit tests in ProfileExporterTests.cpp:

• empty input → empty list
• application id only → 1 entry (regression for the existing rum.application_id emit)
• 1 session, 3 sessions → N+1 entries (multi-value-facet shape)
• empty applicationId → only session entries
• empty session ids in the list are skipped

Manual on Windows (r1viollet):

• Scenario 5 single run: pprof shows 2 rum.session_id:<uuid> entries — one auto-gen'd, one 99999999-...
• Scenario 5 with 11 forced extra rotations (now removed from the runner): pprof shows exactly 10 rum.session_id:<uuid> entries; profiler log contains rum.session_id tag cap reached (10); N session id(s) from this export not emitted as tags

🤖 Generated with Claude Code

[r1viollet]

We discussed this.
This will help with long term approach. I think we can still consider it.
Short term we can focus more on having metrics associated to profiles to help with the filtering.