Push Ollama to its limits, improve ollama using tool results prompt and instructions to generate tool calls

[CorrM]

These edits show better results.

Here are some situations I encountered:

• Sometimes it doesn't answer my question in the UI and logs show it throwing back the JSON response it received from the tool call.
• Sometimes it suggests me (the user) to call the tools.

Tests were conducted with these models:

• llama3:latest (was not able to make a successful tool call)
• qwen2 (sometimes returns raw JSON results or is unable to perform multi-tool calls)
• deepseek-coder-v2:latest
• mistral:latest

Note: qwen2 is the best model depending on my tests.

[CorrM]

Not ready to merge

[CorrM]

After tests, it's ready to merge. @ysolanky @jacobweiss2305

Do your tests before merge