fix(demo): use the managed analysis runtime

[drewstone]

What changed

• Remove the stale system-Python preflight from the live product demo.
• Record Braid’s bundled uv, Python 3.12, and the pinned agent-eval version.
• Keep CLI Bridge readiness in the parallel startup probe.

Why

Braid owns an isolated analysis runtime. The demo incorrectly required an unrelated system Python installation to match it, so a valid installed package could fail before the product ran.

Direct product use

Installed @tangle-network/braid@0.1.1 was used against CLI Bridge main at 9658c4c with Pi 0.83.0 and pi/tangle-router/glm-5.2. A typed slugify task completed in 54.5 seconds, rendered nine tool calls, wrote the implementation and tests, and independently passed 9/9 tests. /activity showed the retained run, AgentProfile, local execution, provider session, 17,966 input tokens, 2,247 output tokens, seven model calls, and complete history.

Checks

• pnpm check
• 638/640 tests passed; two cloud-credential cases skipped as designed
• 15/15 performance tests
• 2/2 storage throughput tests
• 19/19 live-demo tests
• 15/15 live self-tests
• CLI Bridge adversarial matrix passed
• pnpm typecheck