docs: refresh the claims that no longer match the code

[yusuf-gundogdu]

The documentation carries claims that no longer match the code. This corrects sixty-one of them across twenty-seven files. Every claim was checked against the repository before it was changed, and accurate history was deliberately left alone: sentences like "listed since 0.9.59" or "shipped since 0.9.60" are still true and are untouched.

What was wrong, by weight:

• API_DOCS.md announced version 0.11.0 and listed thirteen engines, missing libSQL, DuckDB and Cassandra. Its storage example also used a collection key, queryHistory, that STORAGE_COLLECTIONS does not define, so a client copying it wrote a key nothing reads. The rate-limit table omitted the three storage routes that share the query bucket.
• ARCHITECTURE.md said fourteen database backends where the SHIPPED record it cites carries seventeen type ids.
• SECURITY.md said six fields are encrypted. connection-secrets.ts marks seven, and the list of what stays readable was incomplete, which is the kind of gap a security review acts on.
• AGENT.md gave three different model counts (twelve, ten) against the twenty-eight in measured-profiles.json, and said nine engines cannot be read by a run where the number is fourteen.
• FEATURES.md listed four agent workflows where AgentRunWorkflowType defines five, and quoted single budget ceilings that are now set per workflow.
• AGENT_DEMO.md told a presenter that only two engines provide a read-only statement path. DuckDB implements it too, so the correct number is three.
• The rest are engine and provider counts, channel states that moved, and page-level corrections under docs/providers/ and docs/llms/.

The roster count in docs/llms/setup.md and docs/llms/methodology.md is not touched here; #632 already fixes it.

Verification: the documentation guard tests pass (api-docs-types, agent-documentation, distribution-check, marketplace-copy: 238 pass, 0 fail; security-check and generate-channel-showcase: 51 pass, 0 fail). The helm chart tests fail identically on main in this environment because the chart dependency is not vendored, so they are unrelated to this change.

One commit per file, so a reviewer can read each correction next to the file it belongs to.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud