docs(evaluation): list the actual DETECTORS identifiers

[joeymussalli]

EVALUATION.md names ten ecosystems in prose and then tells readers, forty lines later, that a scan can be narrowed with DETECTORS. It never says what to write. The root README gives three examples and an ellipsis -- claude_sdk,openai_sdk,google_adk,... -- so anyone wanting to restrict a scan to, say, LangGraph has no documented token and no way to find one short of reading the scanner's source.

That gap has a sharp edge. An unrecognised value is forwarded to the scanner verbatim, so narrowing to a name that does not exist produces a scan with nothing to detect -- which reads as a clean repository rather than as a misconfiguration. This is the same trap EVALUATION.md already warns about under "An empty result is not a pass", arrived at from a different direction.

Added the full identifier table next to the ecosystem list where a reader is already looking, plus two clarifications that are not obvious from the names: langchain covers LangChain and LangGraph, which is why ten ecosystems do not map to ten identifiers; and openshell is a cross-cutting detector for shell-invocation tools rather than an SDK.

Identifiers cross-checked against two sources: the scanner's own documented --detectors list, and the SDK-named top-level directories of the trustabl-rules repository, which agree.

Documentation only; no code changes.