[BUG] Ollama provider: premature context condensing and incorrect tool result handling

[navedmerchant]

Describe the bug

The native Ollama provider has three issues that degrade tool-calling behavior and context management:

1. Premature context condensing: parseOllamaModel sets maxTokens to the full contextWindow. Since getModelMaxOutputTokens reserves 20% of the window for output, this causes the context to be condensed far earlier than necessary.
2. Tool results sent as user messages: convertToOllamaMessages always sends tool results with the "user" role, so the model cannot distinguish tool results from actual user messages.
3. additionalProperties in tool schema: The tool parameter schema includes additionalProperties, which is not part of Ollama's tool schema definition and breaks tool-calling templates on some models.

To Reproduce
Steps to reproduce the behavior:

1. Configure a native Ollama model with a large context window.
2. Use a tool-calling workflow with the model.
3. Observe premature condensing and/or malformed tool-call behavior.

Expected behavior

• maxTokens should reflect max output tokens (not the full context window), inheriting the sane default (4096).
• Tool results should use Ollama's native "tool" role with tool_name when the corresponding tool_use block is known.
• additionalProperties should be stripped from tool parameter schemas.

What version of zoo are you running
latest

Additional context
The fix tracks tool_use IDs to tool names so tool results can be sent with the correct role, inherits the default maxTokens, and strips additionalProperties from tool schemas.