docs: make TODO forward-looking, correct stale claims in README

[mburns]

Docs only — no code changes. Suite still 421 passing.

TODO.md: 308 lines → 57

It had become a log of completed work — nearly every line was [x] — with
only two genuinely open items buried inside it. It's now about what's next:

• In flight — nothing.
• Next up — bearer-token auth for the API (the largest functional gap: /api
  writes need a browser session + CSRF, so no script or mobile client can use
  it), and awards/gold.
• Known limits — a table of things that work but have a documented ceiling,
  each with what changing it would cost.
• Deliberately out of scope — so they aren't re-proposed.

Nothing was lost. The sqlean per-module verdicts it carried already live in
docs/sqlean-plan.md; the schema reasoning is in docs/sqlite-features.md;
the history is in CHANGELOG.md.

Stale claims corrected

Claim	Reality
"317 specs" (TODO ×2)	421
vote counts use votes_post_id_idx	dropped as redundant in [113]
sqlean pinned 0.28.3	0.28.4
CI matrix is 5.1 / 5.4 / luajit (README)	5.1 / luajit / luajit-openresty
CI runs "super-linter"	replaced by actionlint/shellcheck/hadolint
TODO.md is "the living roadmap and changelog of what's shipped"	it's neither now
uuid "minted with sqlean's uuid4()"	stamped at insert by utils/public_id, openssl.rand fallback
API route /api/user/:name/about	:username
docs/sqlite-features.md points at TODO.md for removed-views history	that history is in CHANGELOG.md

The one worth calling out is in docs/sqlean-plan.md (and TODO), which
recorded:

stats / math — not adopted: hot/rising math stays in sort.lua;
revisit if SQL-side ranking becomes a bottleneck.

SQL-side ranking was adopted in #23 — controversial orders by POW. A doc
that records a decision which has since been reversed is worse than one
that's merely out of date, because it reads as current intent. Also fixed there:
uuid was listed as "deferred, for the future API phase"; it's adopted.

README additions

• The three-way split of where things are written down, with a read this
  before changing the schema pointer at docs/sqlite-features.md.
• Cursor semantics: new pages by keyset with no depth limit; ranked sorts
  address the first 1000.
• The full CI picture — the stylua job, the 80% coverage gate, and the separate
  lint workflow — plus a note that the Docker job builds a fresh database, so
  it proves migrations on a new install but never exercises the upgrade path.
• Account deletion in the feature list.

One thing I could not verify

I corrected the lapis migrate instructions: the entrypoint runs lapis server
only, so migrations need a docker exec that repeats the entrypoint's two path
exports (otherwise lapis can't find its rocks). That's read off
docker-entrypoint.sh and the Dockerfile rather than executed — Docker isn't
running in my environment. Worth a sanity check next time you bring the
container up.

🤖 Generated with Claude Code

https://claude.ai/code/session_0128hUpuk1spKzk4UHburdki