Fix LiteLLM provider temperature handling for GPT-5.5

[aronnoordhoek]

Bug Description

When using Hindsight with the LiteLLM provider and Azure OpenAI GPT-5.5, internal LLM calls fail because Hindsight forwards explicit temperature values.

GPT-5.5 only supports the default temperature, so calls with values such as 0.0 or 0.1 are rejected.

Steps to Reproduce

1. Configure Hindsight with:

   HINDSIGHT_API_LLM_PROVIDER=litellm
   HINDSIGHT_API_LLM_MODEL=azure/gpt-5.5

2. Run an operation that uses the LiteLLM provider, for example retain/fact extraction.

3. Observe the LiteLLM/Azure OpenAI error about unsupported temperature.

Expected Behavior

Hindsight should work with Azure OpenAI GPT-5.5 through the LiteLLM provider.

For models that do not support configurable temperature, the LiteLLM provider should omit temperature rather than forwarding explicit internal values.

Actual Behavior

The LiteLLM provider forwards explicit temperature values from internal calls.

Examples include:

• connection verification uses temperature=0.0
• retain/fact extraction uses temperature=0.1
• think uses temperature=0.9
• mental model delta operations use temperature=0.0

Azure OpenAI GPT-5.5 rejects these calls with an error like:

Unsupported value: 'temperature' does not support 0.1 with this model. Only the default (1) value is supported.

Version

0.8.3

LLM Provider

Other