Release 3.10.0: changelog entries for the last fixes

[erikdarlingdata]

Adds the changelog entries for the 3.10.0 PRs that merged after the first splice (#5548 covered everything through #5536). No issue. No dev to main PR, no tag, no release (Erik's ruling).

Why

The first splice stopped at #5536 (08:05Z). Eight more PRs merged afterwards with changelog text in the pending buffer, and one more (#5552) is still open.

What changes

Eight entries, each appended at the end of its section in docs/changelog/3.10.md (full text) and in the 3.10.0 index of CHANGELOG.md (title-only line). The 3.10.0 count goes from 257 to 265.

• Fixed, in this order: The module map refresh at service start and in the daily sweep reads from its watermark, not two days (#5519) #5528, Custom Views and custom alert rules scope server reads by server_id, not server name (#5525) #5537, Query Store reads on the stored interval tables no longer miss an interval that began just before the table's floor (#5541) #5543, A quiet server's job history and default trace watermark read tries the last week before its whole slice (#5515) #5532, Release checks: small fixes from the final walk of Lite and the Darling Viewer #5542, Lite's Job History reads only the selected time range, and tab loads read off the UI thread (#5457) #5553, then The startup sweeps stop a continuous aggregate's jobs before they drop it #5552.
• Added: Lite's Job History reads only the selected time range, and tab loads read off the UI thread (#5457) #5553 (the DuckDB memory limit setting).
• [#N] ref links added for The module map refresh at service start and in the daily sweep reads from its watermark, not two days (#5519) #5528, A quiet server's job history and default trace watermark read tries the last week before its whole slice (#5515) #5532, Custom Views and custom alert rules scope server reads by server_id, not server name (#5525) #5537, Release checks: small fixes from the final walk of Lite and the Darling Viewer #5542, Query Store reads on the stored interval tables no longer miss an interval that began just before the table's floor (#5541) #5543, The startup sweeps stop a continuous aggregate's jobs before they drop it #5552, Lite's Job History reads only the selected time range, and tab loads read off the UI thread (#5457) #5553 in CHANGELOG.md. split regenerated the archive's own link block.
• tools/changelog/archive-census.txt: the 3.10.0 row is now 265 entries with the new hash (written by census).

How: I added the full entries to docs/changelog/3.10.md by script, put the ref links in CHANGELOG.md, and then ran split. I did not put full entries into the CHANGELOG.md 3.10.0 section. split takes a version's prose from its existing archive. It refuses, or drops text, when both files carry prose. split then wrote the title-only index lines itself. Of the 12 files split writes, only these three differ from dev.

Entry text is the buffer text as written, with one wording fix from the seat in the #5553 Added entry ("and so does the log" in place of a stray ",."). Entries #5528, #5532, #5537, #5543 and #5552 use ([#N]). text rather than ([#N]) - text, as do 14 existing 3.10 entries. I left those as they are.

#5552's entry comes from the h3 file. Its REF line cited #5551, which no existing 3.10 entry does. So only the [#5552] ref link was added.

#5552 is not merged yet. Its entry is here, and its PR must merge before this one.

Proof of the entry lines

Script (worktree scratch, not committed) took the (section, line) set from the buffer and the h3 file, and compared with what the files gained.

• docs/changelog/3.10.md gained 8 entry lines, lost 0. Sorted-line SHA-256 ada738f0ddb0579deb20de78f8af595aad9e2f3b003971937be0c253e60512da, identical to the SHA of the 8 expected lines (the buffer lines with the seat's one-sentence fix applied to the Lite's Job History reads only the selected time range, and tab loads read off the UI thread (#5457) #5553 Added entry).
• CHANGELOG.md gained 8 index lines (plus 7 ref links), lost 0. Sorted-line SHA-256 26683e3f0976182caa1aadba1fd9da5e2f8cfb9a83b6d201663009645b2a33d2, identical to the SHA of the title-only form of the 8 archive lines.
• Byte census after writing: both files CRLF throughout in the working tree, 0 bare CR (CRLF count equals LF count equals CR count).

Accounting: every PR merged after 08:06Z

PR	Record
#5528	buffer, Fixed (this PR)
#5537	buffer, Fixed (this PR)
#5543	buffer, Fixed (this PR)
#5532	buffer, Fixed (this PR)
#5542	buffer, Fixed (this PR)
#5553	buffer, Fixed and Added (this PR)
#5540	buffer, ## None
#5546	splice-log line, no entry by design
#5548	splice-log line, the release prep itself, no entry
#5550	splice-log line ("no changelog entry")
#5552	entry from the h3 file. Merges before this PR (still open when checked)

Every merged PR in the fresh list (10 of them: 5553, 5550, 5548, 5546, 5543, 5542, 5540, 5537, 5532, 5528) has a record. No PR is missing one.

Test plan

• changelog_archive.py split (twice, second run idempotent: "wrote 12 files"), census (wrote archive-census.txt).
• changelog_archive.py verify: every check PASS. The last line is "PASS ... 3.9.0: 760 entries ... matches", and 3.10.md's "274 distinct" refs resolve.
• changelog_archive.py roundtrip: "PROSE INTACT every version 3653e8d held is byte-identical in prose, and every link definition still resolves".
• 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 on Darling.Tests: Total 1400, Failed 0.
• Full suite not run (CI decides).

For the seat to check

• Merge The startup sweeps stop a continuous aggregate's jobs before they drop it #5552 first, then confirm no newer PR needs an entry (fresh list at push time had none after Lite's Job History reads only the selected time range, and tab loads read off the UI thread (#5457) #5553).