docs: the matrix's findings, as filed issues (#198)

[milyin]

Records in the branch map what the matrix has actually produced, now that the findings are tickets rather than rows in a report: #410, #411, #412, #413, #414, plus the measured evidence on #191.

No code change — the point is that the table was never the deliverable. Two of the five carry a diagnosis rather than a symptom, and both came from comparisons no single cell could make:

• cbindgen: an Option<T> by-value parameter decodes by moving out of a raw pointer #412 is narrowed to the by-value decode path, because the same cell with its type declared as a handle compiles. That comparison is what the declaration-policy axis (step 5) was built for.
• cbindgen: a sum payload's drop qualifies std Option into the source module, and leaves the declared type bare #414 gets both halves of one expression's qualification wrong in opposite directions — std Option pushed into the source module, the declared type left bare — which points at a name-plus-prefix construction rather than at the model's reading.

Also worth noting on #191: its premise is now a number that moves with any change (31 of C's refusals and 8 of JNI's arrive as panics), and the matrix surfaced a channel the issue does not name — a deliberate compile-time refusal through a generated assertion, which is correct behaviour but indistinguishable in the report from broken output.