jq reads every file and retains every result before writing pipeline output

With #453, jq -r .id file1.json ... file1000.json | head -n 1 still reads every file and accumulates every evaluated result before output. Non-slurp inputs do not require that collection.

Proposed scope: read bounded ordered file batches and format/emit per-input results using the existing parser and evaluator. Preserve full-consumption results, file order, raw/slurp/join modes, UTF-8 and errors. This is separate from regex issue #435 / PR #556 and from the broader closed streaming PR #155. Byte-stream JSON parsing and incremental evaluation of a single huge JSON value are outside this fix.