Use serverless usage check for auth

[hansent]

Summary

• switch serverless auth from the workspace lookup endpoint to /serverless/usage-check
• cache successful 200 auth results for 1 hour and short-lived 401/402 failures for 1 minute
• add unit coverage for the new Roboflow API helper and the serverless middleware cache/response behavior

Why

Serverless inference should fail as early as possible when an API key is not authorized for inference or when the resolved workspace can no longer spend credits.

Impact

Serverless requests now distinguish unauthorized keys from billing-restricted workspaces, while preserving the existing route skip list and workspace header behavior on success.

Root Cause

The existing middleware only verified that an API key resolved to a workspace ID, which did not catch inference authorization and billing-cap failures early enough for serverless deployments.

Validation

• python -m py_compile inference/core/roboflow_api.py inference/core/interfaces/http/http_api.py tests/inference/unit_tests/core/test_roboflow_api.py tests/inference/unit_tests/core/interfaces/http/test_http_api.py
• python -m pytest ... could not be run in this shell because pytest is not installed in the active Python environment

[grzegorz-roboflow]

I know the _unauthorized_response was already here, but maybe we can move _authorization_error_response outside of check_authorization_serverless middleware (since it's potentially hot path)