Add MIPROv2 optimizer implementation

[conikeec]

Summary

Implements MIPROv2 (Multi-prompt Instruction Proposal Optimizer v2) as an alternative to COPRO for prompt optimization. MIPROv2 uses LLMs to generate program descriptions and candidate prompts based on execution traces and prompting best practices.

Key Changes

• Core optimizer (mipro.rs): 3-stage process
  • Stage 1: Generate execution traces from training data
  • Stage 2: LLM generates program description + candidate prompts using prompting tips
  • Stage 3: Evaluate candidates and select best
• 29 comprehensive test cases covering traces, selection, configuration, edge cases
• Example (08-optimize-mipro.rs): Full working example with HotpotQA
• Prompting tips library: 15+ best practices built-in
• Documentation: Implementation guide and usage examples
• README updates: Optimizer comparison and configuration details

Implementation Details

• Follows Rust best practices (no unsafe, proper error handling, async throughout)
• Strong typing with Trace, PromptCandidate, PromptingTips
• Builder pattern for flexible configuration
• All tests passing

Trade-offs

MIPROv2 trades speed for quality - makes more LLM calls but produces better prompts by leveraging prompting techniques.

Use MIPROv2 when:

• Complex reasoning tasks
• Have decent training data (15+ examples)
• Quality matters more than speed

Use COPRO when:

• Need fast iteration
• Limited compute budget
• Straightforward tasks

Testing

cargo test optimizer::mipro::tests
cargo run --example 08-optimize-mipro --features parquet

[krypticmouse]

LGTM! I'll suggest to store all tests in the test folder and docs in the suitable group as well! But now worries if you can't I can take care of it!

[krypticmouse]

Thank you so much for the contribution!

[krypticmouse]

As another suggestion we can also use Prompting tips as output from Signature but it's mostly an implementation choice.