The batch commands disagree about exit status, and upload --json emits two documents #27

Symptom

The three batch commands disagree with each other about exit status, and one of them emits a
--json stream that cannot be parsed. Three related defects:

1. assets edit exits 0 when every edit was rejected. It prints Edited 0 photographs. and
returns Ok(()). upload and download both bail! in the equivalent case. A script that
submits 100 edits and has all 100 refused by the server sees success.

This one directly undercuts #23: that issue made the rejection say which field was wrong, but
a caller checking $? still never learns anything was wrong at all.

2. upload --json emits two JSON documents when a file fails. ctx.out.json(...)? writes the
result, then bail! sends main::report to write {"error":"N files failed",…} to the same
stdout. The stream is then two concatenated documents, which breaks jq and any ordinary parser.
src/output.rs states the contract this contradicts.

3. download --json exits 0 while the identical run in prose exits 1. Its JSON branch returns
before the bail!. The mirror image of (1) — and it means the same failure reports differently
depending on which audience asked.

Reproduction

1. imogen assets edit <id> <id> --taken-at <value the server refuses>; check $? — 0.
2. imogen upload <dir> --json with one file the server rejects; pipe to jq — parse error on
   trailing content.
3. imogen download <selection> --json where a file fails; check $? — 0. Run the same without
   --json; $? is 1.

Expected

One rule, applied by all three: a batch in which every item failed exits non-zero; a partial
failure exits with whatever the commands agree on (and they should agree); and --json emits
exactly one document on stdout whatever happens, with the failure represented inside it
rather than appended after it.

Worth settling the partial-failure case deliberately rather than per-command, since that is how
the three drifted apart.

Affected repos

Only this one. Nothing on the wire changes.

Context

Found by review while implementing #23 and #24, both of which were about the content of a
failure report. This is about its exit status and framing, which is the half a script actually
depends on. Pre-existing in all three commands.