This mostly affects nMigen.

This is not explicitly described in the SVF specification, but is
relied on by real-world tools.

In RECV-COUNT-LSB, this used to overwrite the high bits of count.

Fixes GlasgowEmbedded#76.

Such flashes return 0xff as short manufacturer ID, and we might
as well display the long manufacturer ID.

Fixes GlasgowEmbedded#91.

Unfortunately, this takes liberties with the FX2 interface; the data
is now captured on falling edge, so instead of one full cycle of
data validity there is only a half cycle. However, the FX2 interface
was certainly broken before, and it seems less broken to me now, so
I am committing this anyway.

Ideally, the FX2 interface would register the input on the next
rising edge. However, this would add two registers in the input path,
and we are already using the INFM1 bit in the FX2. Therefore, adding
this register would require a radical rethinking of the entire FX2
interface.

See GlasgowEmbedded#89.