Trace: end-to-end test matrix for agent harnesses #3635

Chainloop Trace supports several coding agents (Claude Code, Cursor, OpenCode, and Pi soon). Each agent supports a different set of Trace features. The trace README lists them in a feature table.

Nothing checks this table against the real agents. An agent upgrade or a model change can break a feature, for example the welcome message or spec capture, and nobody sees it until a user reports it.

Add an end-to-end test framework that runs each supported agent headless with real models, checks each feature the table marks as supported, and compares the result matrix with the expected one. A periodic job runs the suite and reports when a cell changes from passing to failing. Features the table marks as not supported are not tested.