docs: tell users a malformed UUID in a multi-row batch declines the whole batch (validate client-side or use a String column) #723

Follow-up to draft PR #712 (branch chtypes-v2).

Problem. ClickHouse's UUID reader swallows the records after a malformed UUID: text shorter than 36 characters makes it take a fixed 36-byte window and resume at the end of that line. Measured on a 26.8 build: [{"page":"/a","id":"x"},{"page":"/b"},{"page":"/c"}] published 0 rows behind a 200, and a server running with input_format_allow_errors_ratio stores the same subset, so this is server behaviour. #712 therefore fails closed: when WaveHouse's own record count exceeds chtypes' verdicts, the whole batch is declined (fail-closed by design).

Docs ask. The ingest and type docs should tell users, in an ingest and a column-types section:

• a single malformed UUID declines the whole batch, not just its record;
• validate UUIDs client-side, or declare the column String if input may be malformed.

Waiting on chtypes to report the swallowed records individually (tracked upstream in chtypes); the guidance can be revisited then.