Batch — run up to 20 calc sub-requests in one call

Runs up to 20 calculation sub-requests in a single call, in order, returning one result per sub-request. Each sub-request is dispatched through the same authenticated, metered handler as a direct REST call (so it records one usage event and is validated identically), and a sub-failure (bad path, bad body) is isolated — it does not fail the batch, which always returns 200. Sub-request paths must be calc routes (no /v1/batch nesting, no /v1/usage or /v1/signup). Rate limiting is all-or-nothing: the batch reserves one token per sub-request up front and is rejected whole with 429 if the budget can't cover them. The whole batch shares one calc deadline budget. Deterministic — identical batches produce an identical result_id.