docs(specs): record why the interactor overlay is built the way it is

[adamjohnwright]

The reasoning behind the interactor work lived only in commit messages. This puts
it where someone will find it.

specs/002-interactor-overlay-decisions/ records eleven decisions, each with
what was rejected and the measurement that settled it:

	Decision	Settled by
D1	Counts are interactions, not entities	Reactome-FIs read 15 beside a badge reading 17
D2	A diagram tally counts each protein once	15 badges over 13 accessions
D3	No badge below 0.6 zoom	6 screen pixels at the 0.283 a pathway opens at
D4	Say so when they cannot be seen	9 badges, none visible, nothing said
D5	The download is the data, not the view	a 12-row file beside a badge reading 73
D6	Threshold remembered per resource	the old browser's own map
D7	A gesture and a replay are different things	two released bugs from one cause
D8	A reader's data is read in the browser	323 stored uploads, oldest 2019, no expiry
D9	Nothing asked until the panel opens	14 third-party requests per pathway → 1
D10	The control is called "Interactors"	two readers missed a working feature
D11	Header controls stay divs, made operable	focusable: false

Plus the custom upload contract, measured by probing rather than read from
documentation that does not exist: the header the service wants, the third column
it refuses, and the first pair it silently eats without a header.

Two things beyond the template

• Every figure is attributed to a measurement and a date. Several of these
  decisions exist because a measurement contradicted an assumption, and a reader
  should be able to tell which.
• Open decisions are recorded as open — where the control belongs, and the
  badge colour — rather than omitted. An absent decision reads as an oversight; a
  recorded one reads as a choice nobody has made yet.

Documentation only; no code changes.

🤖 Generated with Claude Code