chore(cleanup): delete dead eval-persona facade + orphaned topology module

[drewstone]

Removes two confirmed-dead runtime surfaces and bumps to 0.76.0.

Deleted

• evalPersona / EvalPersonaOptions (src/conversation/eval-persona.ts + test) — the one-call user-sim facade, superseded by the brain-from-profile overhaul (refactor: usability overhaul — brain-from-profile, surface shrink, docs that can't lie, supervise() one-call #347). Zero external consumers (grep'd ~/code excluding node_modules + every agent-runtime checkout). The in-repo examples/product-eval is migrated to the kept lower-level runPersonaConversation seam, so the example stays live.
• src/topology/ — createTopologyView / renderTopologyTree / createReplayRecorder / renderReplayHtml. @experimental, no package.json export, no consumer. bench/src/atom-humaneval.mts drops only its optional HTML replay side-output; the driver-vs-blind verdict is untouched.

Kept (re-verify found a real consumer)

• runPersonaConversation / runPersonaDispatch / PersonaDriver (src/conversation/run-persona.ts) — tax-agent (tests/eval/self-improve.ts, dep ^0.75.1) imports runPersonaDispatch + PersonaDriver from the published package. Per the re-verify-before-delete rule, the persona loop runner stays.

Verification (all green, clean worktree off origin/main)

• pnpm run build — DTS emitted for all entries
• pnpm run typecheck — 0 errors (incl. examples)
• pnpm test — 112 files, 1102 passed / 1 skipped / 0 failed (the closed-loop lifecycle proof + full suite green)
• pnpm run lint — 319 files, clean
• pnpm run docs:check — docs:api regenerated, docs/api diff committed, freshness gate OK (version pin 0.76.0 accurate, all citations + symbols resolve)

Net: 16 files, +38 / −1120.

[tangletools]

✅ Auto-approved PR — a56ba63c

Blanket team auto-approval is enabled for this reviewer service.
The full PR reviewer audit still runs separately and will publish findings if it detects issues.

_{tangletools · auto-approval · reason: blanket_auto_approve · 2026-06-23T07:59:41Z}