fix(monitoring): scope heatmap color intensity to the agents actually shown

[pedrofrxncx]

Follows #5865 (tool-call heatmap by agent), which just merged.

In ToolAgentHeatmap (Overview tab), the color-intensity scale (maxValue) was computed over all (agent, tool) cells that use a top tool — before narrowing down to the top-N agents actually rendered. An agent that doesn't make the top-N by overall total, but has one large spike on a single top tool, would still inflate maxValue, silently dimming every cell of the agents that are shown (their true max never reaches the top color bucket).

Why a maintainer wants this: the heatmap is a proxy for spotting the heaviest tool/agent combos at a glance — with the bug, the color scale can be stretched by data the user never sees, making the visible cells look uniformly lighter than they should.

Fix: extracted the tool/agent selection + color-scale calculation into a pure computeHeatmapView() helper (apps/web/src/routes/orgs/monitoring/utils.ts), and moved the maxValue computation to run only over the cells for the agents that end up in topAgentIds. Added utils.test.ts with a regression test reproducing the bug (an excluded agent's spike no longer sets the max) plus a basic top-N capping test.

How to confirm: bun test apps/web/src/routes/orgs/monitoring/utils.test.ts.

Locally ran: bun run fmt, cd apps/web && bunx tsc --noEmit, bunx oxlint on the touched files, and the targeted test above — all clean. Full CI validates the rest.

---

Summary by cubic

Fixes heatmap color intensity in Monitoring Overview by basing the scale only on the agents shown. Visible spikes now render with the correct intensity.

• Bug Fixes
  • Added computeHeatmapView in apps/web/src/routes/orgs/monitoring/utils.ts to select top tools/agents and compute maxValue only for rendered cells.
  • Updated ToolAgentHeatmap in overview.tsx to use computeHeatmapView; resolved merge conflict in overview.tsx after syncing with main.
  • Added utils.test.ts with a regression test (excluded-agent spike no longer skews scale) and a top‑N capping test.

^{Written for commit 2401bbc. Summary will update on new commits.}