Decide paired-seed resampling if seeds are shared experimental units

[esaran1]

Why

resample_runs_by_cell currently draws seeds independently inside each (intervention, compute) cell. That is correct if cells are independent replicates. It is wrong if seed 0 is one paired experimental unit reused across recipes or scales.

What to do

1. From the recipe card (Define the actual recipes and what compute=1,2,4,8,16 means #4), state how seeds are assigned.
2. If seeds are paired, implement a bootstrap that resamples the seed once and applies it across the paired cells.
3. Add a test with a constructed paired table where the independent-cell bootstrap and the paired bootstrap disagree.
4. Document which bootstrap asla audit uses in NOTES_ESTIMAND.md and the audit JSON.

Acceptance

• Pairing policy is written down
• Implementation matches that policy
• A regression test locks the chosen resampling unit
• Audit metadata names the resampling unit

Files

• asla/analysis/audit.py
• NOTES_ESTIMAND.md

Do not fabricate

• Change the bootstrap after seeing the real-data intervals