[capisoft-lib]

Fixes switching from Today to All leaving today's figures visible after a failed history request, and reduces full-history database round trips.

• Keep stale period data hidden behind an explicit error state until the requested period loads. Drop obsolete ETags when a response is no longer cacheable.
• Return completed history summaries without caching when concurrent ingestion advances the revision, instead of discarding the result with HTTP 409.
• Read aggregate pages in batches of 1,024 (raw detail reads remain at 128), select each page before aggregation, and fetch independent metadata concurrently.

Built on current main, preserving the already-merged project grouping, title fallback, historical cache and loading-region changes. The functional and performance changes were deployed in private Sites version 44 and tested by the owner.

Validation: regression coverage for Today → All failure/recovery, ETag invalidation, concurrent ingestion, and complete pagination over 2,050 synthetic sessions. Local before/after comparison preserved the entire response while reducing history reads from 70 to 9; local timings are not a production latency guarantee. Validation completed locally: npm run check (302 tests), hosted build and 15 Sites tests, ESLint, TypeScript, and a clean diff/privacy scan. GitHub CI also validates Windows and the Docker storage smoke test before merge.