Test matrix for streaming, multimodal input, and reasoning controls

[FifoCodeDev]

The current quickstart covers the happy path. Before the next tagged release, I want a small repeatable test matrix for the parts that tend to break when a model or API route changes.

Cases to cover

• normal non-streaming response
• streaming completion and final usage data
• multimodal input
• thinking enabled and disabled
• unsupported parameter handling
• invalid model, authentication, and rate-limit errors
• one fixed prompt set with latency and accepted-output notes

This should stay small enough to run manually before a release. It is not intended to produce a benchmark score.

For expected model behavior, use the official GLM-5.3-Flash announcement as the product source. For the route used by this repository, verify the current model ID and availability on the CometAPI GLM-5.3-Flash model page.

A useful first contribution would be a table of expected response fields for streaming and non-streaming calls.