Prompt update for property extraction

[baitsguy]

An attempt to get better json-parseable responses from the LLM with prompt tweaking. In this case we were previously getting python style dict responses.

Current, non-parseable responses with the json lib:

...
"someFieldWithNoValue": None
...

as opposed to this, which is parseable:

...
"someFieldWithNoValue": "None"
...

Another approach is to get extract_json to better parse these responses with regex/replace style approaches but that felt jankier and I want to avoid that as long as I can

[Soeb-aryn]

LGTM, do you think it is worth adding an assert statement to ensure fields is of right data type ?

sycamore/lib/sycamore/sycamore/transforms/extract_schema.py

Line 209 in 52e1c0b

value = fields.get(field.name)

[baitsguy]

LGTM, do you think it is worth adding an assert statement to ensure fields is of right data type ?

sycamore/lib/sycamore/sycamore/transforms/extract_schema.py

Line 209 in 52e1c0b

value = fields.get(field.name)

The current transform actually seems to work with a dictionary type response, but also just string style responses. I don't think the latter is necessarily needed but let me think about it outside this PR's context