docs: complete framework coverage matrix - #147

Merging, and this closes #96, which I filed in August because TRACE adapter coverage had not moved with the framework landscape. Good to close it with a measurement rather than by deciding it no longer matters.

The distinction between first-party observation and telemetry transcription is the part with lasting value. Those two produce records that look alike and mean different things: one is a runtime describing its own execution, the other is a transcription of what a telemetry pipeline reported about it. Collapsing them in a coverage matrix would overstate what several adapters establish, and §3.1.1's origin block exists for exactly that boundary.

Documenting Pydantic AI's released interoperability and evidence limits in the OTel GenAI README, rather than listing it as covered and leaving the caveats to be discovered, is the same instinct one level down.

Verification is stated and checkable: 24 repository validation tests, 51 OpenAI Agents SDK and OTel/Pydantic adapter tests, the generated integration index and Marketplace catalog regenerated rather than hand-edited, and compatibility checks. Regenerating the index rather than editing it is the thing that keeps validate green repo-wide.