feat: add sference batch inference provider

[apejcic]

Summary

• Add sference as a batch inference provider for European OpenAI-compatible inline batches (POST /v1/batches).
• Extend OpenAIProvider with sference-specific inline submission, polling, and result decoding (result_json / error_json rows).
• Batch only /v1/chat/completions — sference's batch executor rejects Responses-shaped bodies at execution time.
• Include example, generated docs, registry/contract tests, and architecture notes.

Test plan

• pytest tests/test_provider_registry.py tests/test_provider_contracts.py
• Live batchify against https://api.sference.com with moonshotai/Kimi-K2.6 (chat completions path)
• CI green on upstream

Made with Cursor

[apejcic]

hi @vienneraphael. I was playing around with batchling (love it) and validated it against sference (I'm a cofounder).

1. would love to talk to you 1:1
2. would be honored to have sference as batch provider in batchling

[vienneraphael]

hi @vienneraphael. I was playing around with batchling (love it) and validated it against sference (I'm a cofounder).

1. would love to talk to you 1:1

2. would be honored to have sference as batch provider in batchling

Hi @apejcic, thanks for your PR I really appreciate you contributing to batchling through implementing sference!
Sure let's meet, I just sent you a DM on LinkedIn.

PR looks good to me, can you just make sure to run the ruff-check autofix? This should handle the CI issue.
I see some other CI issues not related to the PR, which I'm going to fix on main directly, you should be able to rebase soon after and the CI should go green.

[vienneraphael]

CI fix was just merged on my end, I'll let you rebase and run ruff-check autofix, the CI should work after that.

[apejcic]

Thanks for fixing CI. Fixed formatting with ruff-check. Waiting for CI workflow approval.

[apejcic]

Fixed line width format cf17bd27cca2b7c81127bd40b711a2cf0b7ff1a5

[vienneraphael]

Awesome thank you, merging the PR right now!