Fix #1128 review defects before v3.0.0 tag

[erikdarlingdata]

Pre-tag code review (3 subagents) of this release's alert-badge and XML-processor work surfaced six fix-before-release defects. All fixed and verified in this PR.

Lite server-tab badge parity (#1128)

• Stale badge never cleared. The low-disk / failed-job badge flags were only written inside the feature-enabled / online / msdb branches, so a disabled feature or an offline server left a previously-lit badge stuck on. Now computed as locals through the sweep and written once at the end — the off/offline path clears it.
• Dismissed badge could never re-show. The low-disk and failed-job conditions are plain booleans with no event timestamp, so UpdateAlertCounts' timestamp-based ack-clear never fired for them — once dismissed, a fresh breach / new failed job stayed dark. Added AlertStateService.ClearAcknowledgementForNewCondition, called on a false->true transition, matching the Dashboard's re-show on a new disk/job condition.
• Dict leak / stale flash on reopen. Per-server badge state wasn't dropped on tab close. Now removed in CloseServerTab.
• Tests. New AlertBadgeAckTests (4 tests) cover ack → suppress → new-condition re-show, the no-op-when-nothing-acked path, and silenced-server precedence.

install/25_process_deadlock_xml.sql

• Perpetual NO_RESULTS. The proc logged NO_RESULTS + "left unprocessed for retry" after it had already marked the rows processed — a permanent false-failure signal. A clean parse run with 0 reconstructable graphs is SUCCESS (events processed, simply no deadlock graph). Genuine failures still take the CATCH path → ERROR. Mirrors process_blocked_process_xml.
• Window pad. Moved the +1s pad from @end_date onto the parser's local upper bound only; the mark uses the un-padded UTC @end_date, so a deadlock inserted in that extra second is left for the next run rather than marked unparsed.

install/53_collect_server_properties.sql

• Azure tier/SLO change missed. SERVERPROPERTY('Edition') is the constant 'SQL Azure' on Azure SQL DB, so a pure tier/SLO change with no vCore/memory delta never changed the row_hash and the collector skipped it. Added DATABASEPROPERTYEX(DB_NAME(), 'Edition'/'ServiceObjective') to the hash.

Verification

• Lite build OK; Lite.Tests 447 (443 + 4 new), Dashboard.Tests 487 — all green.
• Installer applies 55/55 scripts on SQL2017, 46 collectors ran (validates modified 25 + 53 in the real install path).
• Diff is surgical (86/28 across the 4 modified files), no line-ending churn.

🤖 Generated with Claude Code