When reading the BLIF input, represent the native sequential elements
with fine-grained cells like `$_FF_` instead of the coarse-grained cells
like `$ff` which we were using up to now.
There are two reasons for this:
* The sequential elements in BLIF are always single-bit, so the gate
cells are a better fit.
* This makes it symmetrical to the BLIF backend which only understands
the fine-grained cells, and only translates those to the native BLIF
features.