docs(plugin-api): make the §8 exports table exhaustive (#457)

[Aybook]

Closes #457.

docs/PLUGIN_API.md §8 closed with "See each plugin's source for the full exported surface", so it read as a contract but was a 7-row sample. In #447 PR 6 that made "not in §8" look like evidence of a dead export and cost a full re-derivation. This makes §8 the exhaustive enumeration it reads as.

What changed (re-derived from source, not from the issue)

• 19 plugins return a top-level export table; 15 have a public (bare-name) surface, 4 (cmd_gag, etc_blocklist, etc_whitelist, hub_runtime) export only _-prefixed unit-test / migration seams. (The issue said "18 exporting plugins" and missed cmd_gag / etc_whitelist / hub_runtime.)
• 4 of the 7 previously-listed rows were themselves incomplete: bot_opchat missing bot, cmd_ban missing del/bans_path, etc_hubcommands missing has/list, etc_report missing broadcast.
• Live doc bug fixed: cmd_ban exports bans as a table by reference (mutated in place, per the etc_msgmanager.tbl re-read from disk on every accinfo / GET /v1/registered/{nick} #238/cmd_ban exported bans table goes stale after cleanbans() rebind #239 idiom), not the callable ban.bans() the old table showed. Verified against importers (cmd_accinfo.lua:198 local bans_tbl = ban.bans, etc_prometheus.lua:111 pairs( ban.bans )). New row documents ban.bans + the ban.bans_path load-on-demand alternative.
• Intro now states the completeness contract and the _-prefix convention, so absence from §8 provably means "exports nothing public" - the exact ambiguity that burned Dead-code audit: findings + cleanup tracker #447 PR 6.

Deliberate altitude decision (so it is not read as an oversight)

The fact-check pass flagged that three exports carry extra optional impl params the plugin author does not advertise: cmd_ban.add's 6th permanent arg (absent from cmd_ban's own export comment even post-#444), etc_trafficmanager.add/del's trailing user, and etc_hubcommands.add's minlevel/opts. §8 mirrors each plugin's own advertised export contract; these undocumented optionals are left out of a quick-reference table. Documenting permanent here would also couple this docs PR to #444's not-yet-promoted state.

Review (§1a.6)

Two independent read-only reviewers on disjoint axes (fact-check vs source / editorial + consistency), main-thread synthesis. Fact-check: no BLOCKER - completeness, the four-seam-only claim, every method name, spot-checked signatures, and the whole ban.bans-by-reference claim all verified. Editorial: table renders (3 cells/row), #10-common-pitfalls link valid and §10.8 genuinely covers #238/#239, no em/en-dashes. Both NITs folded in: import-column inner-spacing aligned to the doc-wide hub.import( "..." ) convention, and the ban.bans note reworded for clarity.

Docs-only, no code change. 3.2.x only.

🤖 Generated with Claude Code