feat(overview): distinguish failed from successful surveys

[dwolfson]

Problem

A survey that fails still writes a SurveyReport and whatever findings it reached before it throws, so "a report exists" does not mean "this asset was surveyed". On the live platform the PostgreSQL server surveys always threw (OMES-SURVEY-ACTION-0018) yet their reports counted toward Surveyed, quality annotations and schema coverage on the Overview.

Change (quickstart and freshstart)

Each report is classified from its own properties (_survey_outcome):

• engine-run surveys carry a completionMessage: OMES-SURVEY-ACTION-0019 ("has completed the analysis") → completed; OMES-SURVEY-ACTION-0018 ("… threw …") → failed;
• reports created by other tools carry error_count in additionalProperties: 0 → completed, more → failed;
• anything else (e.g. the database surveys that only record table/column counts) → unknown, never counted as a success.

Surveyed assets, surveyed stores / never surveyed, quality-annotated assets, survey-analysed schemas and the findings tally now come from completed surveys only (a failed run's partial findings are excluded). A new Attention Queue row "Surveys that failed" shows failed reports, assets that have only failed surveys, and reports with an unknown outcome, linking to Operations. If the reports cannot be read, the coverage figures are None (not assumed successful).

Live check: of 16 survey reports, 10 completed, 3 failed, 3 had no outcome recorded. Surveyed went to 8 assets, with 2 more having only failed surveys.

Tests

• 23 backend survey tests (outcome classification, subject with both a completed and a failed survey, failed-run findings excluded, lookup failures, caps, the real treasury case) + the 20 existing helper tests, both handlers.
• Page harness: 78 (quickstart) / 72 (freshstart) checks incl. grammar, unknown/None states, call order.
• Real browser (Chrome on the same machine): new row and tooltips render with live data.
• Freshstart not run live.

Notes

• The outcome rule is inferred from two conventions seen on this platform; surveys have no standard outcome field. The database surveys created by Resource Explorer record only counts, so they show as "unknown" until they also record error_count (or the platform gains a completion status on SurveyReport).

🤖 Generated with Claude Code

Follow-up commit: Egeria 6.2 annotation reuse (see #604)

Egeria 6.2 reuses survey annotations, so one annotation can be reported by many survey reports. Two Overview consequences, both handled here:

• Findings are counted per distinct annotation, not per ReportedAnnotation link (a reused annotation would otherwise be counted once per report). An annotation counts if at least one completed survey reports it; one cited only by failed surveys stays excluded. Quality/schema credit is unchanged.
• Links are read in pages. Each report links every annotation it reports (new or reused), so the link count grows with every survey even when nothing new is created. A single 1,000-link page was already at 913 on the live platform; once capped the quality/schema/findings figures would silently go to "not measured". _relationship_rows now ends on an empty page (a full or short page is not the last page), up to 10 pages; hitting the limit marks the result capped, and a failure on any page discards the partial rows (None, never a half-read count). The annotation read also honours the as-of time now.

Behaviour on the current platform is unchanged (no reuse yet; 772 distinct annotations, not capped). 32 survey tests now (adds reuse, multi-page, the page limit, failure on page 2, the exactly-one-full-page boundary).