feat: add LiteLLM as AI gateway provider

[RheagalFire]

Summary

• Adds LiteLLM as a completion provider, giving gpt-cli users access to 100+ LLM providers (Bedrock, Vertex AI, Groq, Together, Mistral, and more) through the LiteLLM Python SDK.
• LiteLLM is added as an optional dependency (gpt-command-line[litellm]).

Motivation

gpt-cli currently supports OpenAI, Anthropic, Google, Cohere, Azure, and LLaMA. LiteLLM lets users access any additional provider without a new provider implementation per SDK. Users prefix the model name with litellm: and use LiteLLM's model naming convention (e.g., litellm:bedrock/anthropic.claude-v2, litellm:together_ai/mistralai/Mixtral-8x7B).

Changes

• gptcli/providers/litellm.py - New LiteLLMCompletionProvider implementing CompletionProvider
• gptcli/assistant.py - Registered litellm: prefix in get_completion_provider()
• pyproject.toml - Added litellm>=1.80.0,<1.87.0 as optional dependency

Implementation notes

• drop_params=True on every call so provider-unsupported kwargs (e.g., temperature on reasoning models) are silently dropped
• litellm is lazy-imported inside the complete() method so the module loads without litellm installed
• Follows the same pattern as existing providers: yields MessageDeltaEvent for content, UsageEvent for token counts
• Streaming supported via litellm.completion(stream=True)
• LiteLLM reads provider-specific env vars automatically (ANTHROPIC_API_KEY, AWS_ACCESS_KEY_ID, etc.)

Risk / Compatibility

• Additive only. Existing providers untouched.
• Optional extra - base install unaffected.

Example usage

# ~/.config/gpt-cli/gpt.yml
assistants:
  litellm-claude:
    model: litellm:anthropic/claude-sonnet-4-20250514
    temperature: 0.7
  litellm-bedrock:
    model: litellm:bedrock/anthropic.claude-v2

# Set provider-specific env vars
export ANTHROPIC_API_KEY=sk-ant-xxx
gpt litellm-claude

import litellm

# Under the hood, gpt-cli calls:
response = litellm.completion(
    model="anthropic/claude-sonnet-4-20250514",
    messages=[{"role": "user", "content": "Hello"}],
    stream=True,
    drop_params=True,
)
for chunk in response:
    print(chunk.choices[0].delta.content, end="")

[RheagalFire]

cc @kharvd