Bump to 1.38.1

[ww-mw]

A model whose workspace part is cut inside a complex variable opens: MatParser.parseMatrix left the value null for a
complex numeric variable cut before its real part, and MatlabVariableNode._createFromMatNumeric's complex arm then
read .re and refused the whole model; such a variable is now built by the real arms, as a real variable cut at the
same point already is, with no warning (cut every byte over the 22 repo fixtures with a workspace part: 124 of 86,102
cuts refused at 1.38.0, 0 now, 85,978 unchanged, 0 regressed). And the empty-zip check of #114 accepts PK\x05\x06
only as a real empty archive, at least 22 bytes with both entry counts 0, so other bytes that merely begin with it are
refused as Not a Simulink model again, as before 1.38.0 (#116). Both were the "still open" items of 1.38.0 (#117).
Left out, as #116 reports: a workspace variable whose flags word is overwritten (class 255) is damage, not a cut, and
still refuses the model.

Gate: npm run verify passed (rc 0). Test files: 238 passed, 1 skipped (239). Tests: 7922 passed, 26 skipped (7948).
The other stages were also green: typecheck, build, smoke, check:pack, check:leak and check:browser (156 modules, no
node: builtins); dist/ rebuilt from scratch is unchanged.