Release 3.10.0: changelog entries for the last gate fixes

[erikdarlingdata]

Why

Seven PRs merged to dev after splice 2 (#5557). This adds their CHANGELOG records for 3.10.0. No dev to main PR, tag or release.

What changes

• Fixed, appended at the end: the buffer entries for Darling Viewer: Top Queries, Top Procedures and Query Store tabs make fewer store calls (#5555) #5556, Role provisioning runs one batch at a time, and the viewer-grant test replays stop rewriting the shared database's access list (#5560) #5566 and The Query Store interval tables' retention purge deletes in row-capped batches #5572, then seven new entries for Release checks: fixes from the morning walk of Lite, the Darling Viewer and the web dashboard #5554. Changed, appended at the end: Recommendations skip databases that are a secondary copy on this node #5564's buffer entry.
• The Release checks: fixes from the morning walk of Lite, the Darling Viewer and the web dashboard #5554 entries replace its one long PR-body entry. They state the final behaviour by topic, taken from the squash body. The topics are Running Jobs, File I/O with the I/O Latency lane, Lite's own queries, the statement filter on Live Snapshot and saved plans, the Databases filter, server-tab wording with screen readers and Compare, and the web dashboard. The offline fleet card line follows the diff's final tests (it shows the real age of the last collection). The seat buffers them from the entries file after this merges.
• Each full entry went into docs/changelog/3.10.md, its title-only index line into CHANGELOG.md, five new [#N] ref links in each file. archive-census.txt moved from 265 entries to 276.
• Only CHANGELOG.md, docs/changelog/3.10.md and tools/changelog/archive-census.txt are in the diff. Both changelog files stay CRLF in the working tree (no bare CR).

Entry-line proof: the 11 gained lines in each file match the expected set by count and SHA-256 of the sorted lines.

• docs/changelog/3.10.md: 11 lines, 575e5767945033ae1bf9362bdde3b5b67205b3e11a264ed6af9ae3b7f35b82a5
• CHANGELOG.md (index form): 11 lines, e3e29565424d0c09bebe83bfe742d91eefb41f766998e88550c9a38c93142ef2

Accounting for every PR merged since 13:05Z:

PR	Record
#5554	seven new Fixed entries
#5556	buffer entry, Fixed
#5563	no entry by design (test only)
#5564	buffer entry, Changed
#5566	buffer entry, Fixed
#5567	no entry by design (test only)
#5572	buffer entry, Fixed

Every PR in the fresh list has a record. None is missing.

Test plan

• changelog_archive.py split: wrote 12 files, no prose change outside 3.10
• census: wrote the census file (3.10.0 now 276 entries)
• verify: all PASS, including "3.10.0: 276 index lines for 276 archived entries"
• roundtrip: PROSE INTACT, every version the old revision held is byte-identical in prose
• Darling.Tests Debug build: 0 Warning(s), 0 Error(s)
• ChangelogIndexAndArchiveTests: Total 5, Failed 0
• ProductVersionDeclarationTests: Total 26, Failed 0
• ReleaseStatementGateWorkflowTests: Total 4, Failed 0
• -trait Stage=Guard -failSkips: Total 1411, Failed 0
• Full suite not run (CI decides)

For the seat to check

• The Release checks: fixes from the morning walk of Lite, the Darling Viewer and the web dashboard #5554 entries use only squash-body facts, plus the 48-hour offline card line from the diff. Two entries (Running Jobs, web dashboard) carry many short sentences.
• Plain-English checker on handoffs/h2/entry-5554.md reports only dash and "long" hits that come from the house-style - **Title** ([#N]) - line being read as one sentence. Every sentence on its own is 25 words or fewer.