Add silentdrop-llm to Observability (runtime guard for silent LLM-response failures)

[sravan27]

Adds silentdrop-llm under Observability. It catches the silent failure modes of LLM/agent responses — missing required fields, enum drift (e.g. "in-progress" when the schema says "in_progress"), hallucinated IDs that reference nothing in the lookup context, and claim/list mismatches — none of which throw, all of which silently produce wrong downstream data.

Fits next to QWED, EvalView, Giskard, and Fiddler in that section — same lane (runtime validation of LLM outputs) with a different focus: zero-dep, drop-in between the model call and any side-effecting use of the response.

• Repo: https://github.com/sravan27/silentdrop-llm
• npm: https://www.npmjs.com/package/silentdrop-llm
• 7/7 unit tests green, MIT, zero dependencies.

Happy to adjust the section / wording if you'd prefer a different placement.