fix(invoke): encode request body as UTF-8 bytes to prevent Latin-1 corruption

[ekeith (evanmkeith)]

Summary

invoke() passed the JSON body as a Python str to requests.post(data=...),
which encodes str bodies as Latin-1 by default. Any non-Latin-1 character in
the payload (em dashes, smart quotes, etc.) either raised a UnicodeEncodeError
or was silently corrupted before the request was sent.

• Encode bt_dumps() output to UTF-8 bytes before passing to data=
• Add Content-Type: application/json header (was missing; only Accept was set)

bt_dumps is kept — it handles Pydantic models, dataclasses, and NaN/Inf values
that stdlib json cannot serialize. Other SDK paths (logger.py) already use
.encode("utf-8") correctly; this brings invoke() in line.

Test plan

• Unit test: assert data= arg to requests.post is bytes, contains correct UTF-8 encoding of em dash (\xe2\x80\x94), and Content-Type header is set
• Manually run invoke() with Unicode input ("result \u2014 excellent") and confirm no UnicodeEncodeError and payload reaches the API intact

Fixes BT-4620

[Abhijeet Prasad (AbhiPrasad)]

can we add a VCR test instead of mocking if possibe? If it's too annoying no worries.

[Abhijeet Prasad (AbhiPrasad)]

Thanks ekeith (@evanmkeith)!!