`mute_dry`'s Mix-push reach has no test

[bdbarnett]

It had the same see-through-the-wire blindness that two kit tests caught
elsewhere, and it was fixed without a row of its own.

From #91.

[bdbarnett]

Four rows in tests/test_effect_kit_graph.MixPushTest, in 22907a2.

The reach itself: Fuzz's germanium graph blends inside audioshaper.Waveshaper, so there is no voice to mute and the macro is the only handle — the fallback finds Mix, pushes it to 127, and hands back the caller's value to put back.

The planted fault is the see-through-the-wire one this issue is about. A Port's answer to "are you the borrowed source" is always no, because it is a wire and never the source, so the guard has to ask port_target. Blinded to the wire — kit.port_target replaced with the identity, which is exactly what asking the wire means — Exciter at Mix 0 is no longer recognised as a bypass: the clause goes through, pushes Mix to 127, and returns [(_MacroMute, index, 0.0)]. A pack asking for the wet floor of a bypass then gets the floor of Mix 127, with the 0.0 sitting in the return value making it look like a level that was legitimately muted. Measured: the render after the blind push differs from the Mix-0 render, which is the borrowed source byte for byte.

The shipped code refuses, and the row asserts both halves so the guard cannot go quiet again. The other two rows in the same class cover the selector fix from #81.

19 tests green in the file.