fix(analytics): half-open trend buckets to stop boundary double-count

[philluiz2323]

Closes #519.

buildRecommendationQualityReportFromOutcomes builds two time-bucketed views over the same outcomes using the same contiguous trendPeriods boundaries (endMs[i] == startMs[i+1]), but with inconsistent interval semantics:

• trendBuckets: timestamp >= startMs && timestamp <= endMs — inclusive on both ends, every bucket.
• qualityRollups (same feat(analytics): add role-aware recommendation quality rollups #470 PR): >= startMs && (last ? <= endMs : < endMs) — half-open, correct.

So an outcome exactly on an internal boundary was counted in two adjacent trend buckets but one rollup, inflating trends so their sum could exceed totals.total and disagree with the rollup table on the operator dashboard.

Example (generatedAt = 2026-06-01, windowDays = 14 -> split at 2026-05-25): one outcome at 2026-05-25T00:00:00.000Z gave trends = [1, 1] (sum 2) while totals.total = 1 and rollups count = 1.

Change

• Apply the same half-open [start, end) rule (final bucket inclusive) in trendBuckets that qualityRollups already uses.
• Add a regression test asserting a boundary outcome lands in exactly one bucket (trends = [0, 1], sum(trends) == totals.total, and the rollup is in the later bucket).

Verification

• recommendation-quality-report.test.ts 10/10 (the new test fails on the old inclusive-end code). Full unit suite green; changed file 100% stmts/lines, branch coverage holds.

[ghost]

Note

Gittensory Gate skipped

PR closed before full evaluation. No late first comment was created.

Signal	Result	Evidence	Action
Gate result	⚠️ Skipped	#520 is no longer open.	No action.

---

Checked by Gittensory, a quiet PR intelligence layer for OSS maintainers.

[superagent-security]

Superagent didn't find any vulnerabilities or security issues in this PR.

[JSONbored]

@philluiz2323 this is good to land.

A few notes:

• The half-open bucket change fixes the actual boundary double-count in trendBuckets.
• The regression test covers the exact internal-boundary case and checks both trend totals and rollup alignment.
• The missing formal Summary/Safety template sections are nonblocking here because the body explains the bug, change, and validation clearly enough.

No code changes requested.