docs(phi-scan): cut a test-composition sentence that was wrong twice - #39

The third conformance pass on the phi-scan slice found that the sentence describing
which tests stay green under the base scanner under-enumerates: it named "the floor
tests, the controls, and the ones that PIN a residual", and at least 15 of the 38 fall
in none of those categories. They are regression guards from earlier slices: rename
detection, staged non-regular entries, all-mode non-regular entries, and the two
invocation-failure cases. This repo's own vocabulary makes "controls" a narrow
enumerated set, so they cannot be absorbed into it.

That paragraph has now produced a documentation-accuracy defect on three consecutive
passes, which is a signal about the paragraph rather than about the scanner. So the
sentence is CUT rather than restated a fourth time, per this repo's standing "CUT, do
not rewrite" rule. What replaces it is the one fact that is cheap to verify and does
not decay into a taxonomy: 33 of the 38 are tests the base suite already carried,
re-derived here (34 base tests, 33 carried forward, 14 new, 47 total).

The load-bearing number is untouched and was confirmed by the same pass: 9 of 47 red
against cd221a0's scanner. No behaviour changes; this is one markdown paragraph.

A sibling instance of the same pattern one section up, describing an earlier slice's
comparison, is PRE-EXISTING and is deliberately left alone for its own slice.