feat: support annotation output format in harness

[VIDIT45AGARWAL]

This updates the python-jschon harness to support the new output argument.

Changes

• Updated cmd_run to accept the new output parameter
• When output == "annotations", uses jschon's result.output("basic") to extract annotations and transforms them into Bowtie's expected annotation format
• When output == "flag", returns the standard {"valid": bool} response as before

[VIDIT45AGARWAL]

Hey @Julian, I have updated python-jschon harness for annotation support.

[Julian]

Nice! Will have a look (looks like I didn't have notifications on)

[VIDIT45AGARWAL]

It seems smoke tests send a basic payload without the new "output" field, which caused a TypeError. Should we pass a default value output="flag" ?

[Julian]

That would sound like an upstream bug in the main repo -- it shouldn't be possible to send invalid requests to harnesses so if that's happening from smoke something very strange is happening.

We likely should smoke test both kinds of input as well.

[VIDIT45AGARWAL]

I think you are right, this is an upstream issue. The smoke test in Bowtie on main does send output:"flag" , but the harness's report.yml workflow installs bowtie's last release (2026.7.4, Jul 30), which predates the output field on the run command. So its run requests lack output entirely, which is what caused TypeError in the harness.

Also you're right that the smoke test will always just send output:"flag" and not output:"annotations", so we should update that too.