feat: anchor every section heading (A2)

[jackye1995]

The A2 anchor pass, complete. All 810 section headings across the authored pages now carry an explicit {#anchor} — the identity an Enterprise overlay attaches to from A5, and the one thing that survives a heading being reworded.

Stacked on #352; merge that first, then retarget this to main so CI runs (the Assemble workflow only triggers on pull requests to main).

Anchors are read, not derived

They come from the ids Mintlify already renders, read back from a mint export bundle. That is what keeps existing deep links working, and it matters more than it looks — four separate rules would otherwise have had to be reimplemented correctly, and each was found only because the export harness or the alignment guard caught it:

	What happens	Scope
Smart quotes	## What's next? renders as what’s-next with a curly apostrophe. A derived slug gave what-s-next — silently changing the id	1
Numbered headings	### 1. Setup loses its number once an anchor is added; Mintlify re-reads 1. as a list marker and drops it from the heading and the TOC	117 across 17 pages
Setext headings	Text over a rule of dashes gets an id like any heading. A hashes-only parser consumed ids out of order, giving every later heading on the page the wrong anchor	11
HTML entities	An id containing & appears as &; writing that back produced ...-amp-...	5

Headings retitled so their anchors are clean

Eighteen headings across twelve pages produced anchors containing &, /, or curly quotes. Ampersands were the pressing case — Mintlify keeps & in a generated id but strips it from an explicit anchor, so those five could not be anchored at all — but slashes made anchors read like paths, and curly quotes made them non-ASCII. Neither belongs in a key overlays are written against.

The punctuation was incidental in every case, so the headings say the same thing without it:

Observability & performance          ->  Observability and performance
1. Object storage (S3 / GCS / Azure) ->  1. Object storage (S3, GCS, Azure)
What's next?                         ->  Next steps
How is strong/weak consistency ...   ->  How is strong and weak consistency ...

Nothing linked to any of the old anchors, so nothing breaks.

Left alone deliberately: analyze_plan, explain_plan, max_pooling, approx_mode, torch_col, 100B+ row scale. Those characters are part of an API name or a quantity rather than punctuation, they are safe in a URL fragment, and renaming them would misname the thing the heading documents.

Verified

• The twelve pages whose rendered output changed are exactly the twelve retitled — no collateral movement. Every other page is byte-identical to before the pass
• link check clean; the assembler's duplicate-anchor validation accepts the result
• anchors assigned by position, so the script refuses to write anything when source headings and rendered ids disagree — which is what surfaced the Setext and h5 cases rather than shipping them