docs: add experiment strategy tips to program.md

[tykoo-chen]

Problem

AI agents running autoresearch can get stuck or waste experiments on low-impact changes. The current program.md tells agents what to do but not how to do it effectively.

Solution

Add an "Experiment Strategy Tips" section with guidance on:

• High-impact changes to prioritize (LR, model size, batch size)
• Common pitfalls to avoid (changing too many things, abandoning ideas too early)
• When stuck strategies (revert, combine ideas, remove complexity)
• Signs of a good experiment (clear hypothesis, minimal change, interpretable)

This should help agents be more efficient researchers.