Add new langchain agent to support more models

[wbreza]

The default langchain agent is limited to very strict handling of LLM responses which does not work in many models.
Need the ability to adapt for other models.

Acceptance criteria:

• Agent should work well with wide variety of models (Fails for GPT5 for example)
• Agent should do a better job at following directions (Need plan, execute, validation loop within agent itself)
• Agent only sometimes calls tools (best practices, bicep schema, etc)
• Agent uses hard coded "prompt template". Needs to compose better for system messages, etc
  • Prompt is sent as one large "human" message today
  • Needs to be broken down into multiple messages of "system", "human", "assistant" for better internal context handling.

[rajeshkamal5050]

Done. We moved from langchaingo to the GitHub Copilot SDK which handles multi-model support natively. Model selection is available through azd copilot config.