[AI Assistant] Evaluate DeepSeek V4.1 Flash and configurable reasoning

[astandrik]

Goal

Evaluate DeepSeek V4.1 Flash for the AI Assistant and choose explicit reasoning settings based on response latency and answer quality.

Scope

• Compare DeepSeek V4 Flash and V4.1 Flash on the same representative scenarios, including tool calls.
• Compare supported reasoning_effort settings, including low and none.
• Measure time to first visible response and total response time separately; include repeated samples, errors and timeouts.
• Verify answer correctness, tool selection and confirmation behavior.
• Add configurable reasoning settings with focused tests, then document the selected configuration and rollback procedure.

Acceptance

• A reproducible comparison records both latency and quality.
• The outgoing request contains the selected model and reasoning parameters.
• The final configuration is justified by the comparison and has no material quality regression.

Related: #3992, #4366.

References: AI Studio reasoning controls, AI Studio release notes.