docs: strip internal PR/issue numbers and process tells from blog prose

[vivek7405]

Blog posts carried inline internal tracking references (#605, #646, #489, ...) that read as commit chatter to a reader, plus one process tell ("Verified by dogfooding") that leaks the author's verification workflow into published copy. Neither belongs in a blog.

Closes #881

What changed

• Removed every #NNN from blog bodies across 16 posts, rewording each sentence to read naturally (or dropping the sentence entirely where the number was its only content, e.g. "It landed in Elide page/layout modules that ship only to import interactive components #605 and feat: elide page/layout modules that only import components #609.").
• Removed the "Verified by dogfooding: ..." tell from per-action-middleware.md.
• Rewrote the flagged import-only-pages-zero-js.md paragraph so it reads as clean prose.

What was deliberately kept

• The author's rhetorical "Let me show you..." / "Let me start with..." openers. These appear in the original pre-existing posts and are the house voice, not a defect.
• Legitimate #NNN refs elsewhere (agent-docs/, AGENTS.md, docs site) are untouched; only blog/*.md changed.

Verification

• git grep -nE '#[0-9]{3,4}' -- 'blog/*.md' returns nothing.
• No "dogfood"/"self-review"/"counterfactual" tell remains in any post.
• No em-dash or banned pause-punctuation introduced (invariant 11 hook passes).
• Content-only (no src), so the doc/scaffold/test gates do not apply.

[vivek7405]

Read every reworded hunk. The prose reads clean, no dangling clauses or orphaned transitions where a trailing sentence was dropped, and every version number, runtime name, and API name survived the number-stripping. The 'Let me show you' voice is intact. Every removal was a tracking pointer or pure history, not a fact a reader needs. Ready to merge.