[Pipeline B] Wire fill flow to accept input_id and use stored transcript

[abhishek-8081]

Closes #638.

Changes the fill endpoints to accept input_id and use the stored Input.transcript instead
of a raw input_text in the request — the core of Pipeline B's "store the transcript once,
reuse it" model.

• FormFill / AsyncFormFill now take input_id (UUID) instead of input_text.
• Shared InputService.resolve_transcript(session, input_id): looks up the Input via
  get_input, raises 404 INPUT_NOT_FOUND if missing, 409 INPUT_NOT_READY (with the actual
  status in the detail) if status != ready, else returns the transcript. Both routes use it,
  so the rule lives in one place.
• Sync (/forms/fill) resolves after the template check; async (/forms/jobs) resolves up
  front, before the Celery dispatch, so 404/409 return synchronously and no job is queued
  on a bad input_id.
• Contract (form-record.yaml, forms.yaml) updated to input_id (uuid) with 409 examples —
  this is the input_text → input_id switch deferred from the contract-reshape PR, now done
  with the code together.
• FormSubmission still stores the resolved transcript (the FormSubmission → Input FK is a
  separate follow-up issue).

Note: input_id is a UUID (matching the Input model / input-record.yaml), not int as the
issue text said - corrected here.

Please Review @marcvergees @vharkins1 @chetanr25 .