docs: publish bounded cross-vertical paid-agent evidence

[abrichr]

What changed

Publishes the bounded paid computer-use-agent aggregates from merged
OpenAdaptAI/openadapt-flow#216
(exact Flow main e740a7e4b68ed5c6461bdcd6fbd086e8ac75c8be) alongside the earlier compiled/API reference
evidence for the three pinned synthetic environments.

Vertical	Trials	Primary outcome	Over-halt	Silent incorrect	Mean list-price cost
Insurance / openIMIS	3	3/3 correct	0/3	0/3	$0.4793/run
Lending / Frappe Lending	6	6/6 correct writes; 5/6 clean	1/6 post-write cost-cap halt	0/6	$0.4240/run
Healthcare / local OpenEMR	6	0/6 correct; 6/6 missing_write	0/6	0/6	$0.8901/run

The old duplicate cross-vertical section is consolidated, and stale
"agent arm omitted/not run" wording is removed from the current docs.

Dependency and honest scope

• The public aggregate source is merged Flow #216 at exact main
e740a7e4b68ed5c6461bdcd6fbd086e8ac75c8be.
• Paid trials used separately provisioned baselines from the earlier
  compiled/API subsets. They are not matched head-to-head rows or a three-arm
  matrix.
• N is 3-6 per environment. This is local engineering evidence, not a
  publication, certification, broad computer-use-agent claim, customer result,
  or design-partner result.
• All paid tasks used the Browser (Playwright) Beta substrate. They are not
  Windows, RDP, Citrix, or VDI evidence.
• The OpenEMR agent exhausted its action budget without the required effect;
  the oracle classification is missing_write, not over-halt.
• Public docs retain bounded aggregates, methods, failure taxonomy, and
  caveats only. Raw per-run rows, screenshots, fingerprints, detailed cost
  ledgers, and application-specific driver/oracle recipes remain private.

Website/catalog counterpart: OpenAdaptAI/openadapt-web#264.

Exact prepared state

• base: 3c3c71770d84c2969bc7219246647475d2b5c82d
• head: 6a79700bab9c87a346e672f3071e44551166eed9
• commits: 1

Validation

• docs tests: 68 passed
• mkdocs build --strict: passed
• stale current-claim search: passed