docs: how to read the decision log, and the six entries the results depend on

[esamoun]

Adds two sections at the top of docs/decisions.md, above the first entry. No existing entry is edited.

How to read this log. Why 28 of the 60 entries fall between 12 and 14 August (the first end-to-end chain forced a decision at every seam, and each was committed the day it was made), and who did what: the decisions and measurements are mine, the prose was drafted with Claude Code and reviewed, as the README already says under "How this was built".

Start here: six decisions. The six entries the results depend on, one paragraph each, with a link to the full entry and its key number:

1. unsearched rather than dark when no AIS was supplied
2. the study area moved on AIS evidence (19 of 30 Anholt acquisitions had no declared vessel)
3. declarations moved into the radar's frame: 89 of 149 matches, 68.3% vs 21.2% dark without it
4. the ladder rule fixed before training, and R5 rejected by it (−0.0128 against a band of 0.0099)
5. fixed structures excluded by position: 782 of 972 published detections
6. intervals resampled by acquisition: 21.2% [13.6%, 29.4%]

Every number was checked against its entry, and all 8 in-page anchors resolve to an existing heading. No test reads this file.