fix(reports): count only rows that have a value for the selected metric

[JeremieSamson]

Problem

In the performance report, the sample size shown next to a Web Vital does not
describe that metric.

getPerformance and getPerformanceMetrics compute percentiles with
percentile_cont(...) within group (order by <metric>) (Postgres) and
quantile(...)(<metric>) (ClickHouse). Both skip NULL rows. The count next to
them is count(*) / count(), which does not.

Web Vitals are sparse. CLS and INP are Chromium-only, so Safari, iOS and
Firefox pageviews store a website_event row with event_type = 5 and no
value for those columns. The two numbers therefore describe different
populations.

This is visible in the UI, which labels the number sampleSize:

{t(labels.sampleSize)}: {formatLongNumber(data.summary?.count || 0)}

Evidence

Self-hosted instance, Postgres, one website, 14 days, event_type = 5:

Metric	Rows with a value	Reported sample size
TTFB	1681	1681
FCP	1483	1681
LCP	1300	1681
CLS	345	1681
INP	112	1681

So the report claims an INP p75 of 88 ms over 1681 samples, when 112 rows
carry an INP value. The smaller the coverage, the more confident the number
looks.

Fix

Count the selected metric column instead of every row:

count(${metric}) as count

Same change on both engines, in getPerformance.ts (the summary the UI
labels "sample size") and in getPerformanceMetrics.ts (the count returned
for pages, titles, devices and browsers).

Verified on the same data:

 inp_p75 | count_before | count_after
---------+--------------+-------------
      88 |         1681 |         112

Notes

• metric is validated by z.enum(['lcp','inp','cls','fcp','ttfb']) in
  schema.ts, so interpolating it here is no different from the existing
  percentile expressions in the same queries.
• The Web Vitals columns are nullable on both engines, so count(column)
  excludes missing values on ClickHouse as well as on Postgres.
• The UI already re-queries when the selected metric changes (metric: selectedMetric is part of the query parameters), so the summary count
  follows the selection with no UI change.
• No existing tests cover these two query files.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[vercel]

@JeremieSamson is attempting to deploy a commit to the Umami Software Team on Vercel.

A member of the Team first needs to authorize it.

[greptile-apps]

Greptile Summary

This PR aligns reported Web Vital sample sizes with the non-null observations used to calculate percentile values.

• Replaces total-row counts with selected-metric counts in the performance summary query.
• Applies the same counting semantics to grouped performance metrics.
• Keeps PostgreSQL and ClickHouse implementations behaviorally aligned.

Confidence Score: 5/5

The PR appears safe to merge, with no actionable correctness, security, or compatibility issues identified.

The selected metric is constrained to known nullable Web Vital columns, and both database implementations now count the same non-null population their percentile aggregates use.

Important Files Changed

Filename	Overview
src/queries/sql/reports/getPerformance.ts	Counts non-null values for the selected Web Vital in both PostgreSQL and ClickHouse summary queries, matching percentile populations.
src/queries/sql/reports/getPerformanceMetrics.ts	Counts non-null selected-metric values in grouped performance queries across both supported database engines.

_{Reviews (1): Last reviewed commit: "fix(reports): count only rows that have ..." | Re-trigger Greptile}