Switch FPGA API's to be dynamically loaded (Try 2)

[ThadHouse]

I realized I went completely overboard with the API changes in the other PR. I'd still like to look into those, but this just changes to the dynamically loaded API without changing anything else, and no API changes at the high level.

[auscompgeek]

I was wondering when you'd be done with the other PR 😅

My main concern is around the dlopen error newtype, but looks reasonable otherwise.

[ThadHouse]

This should now not be any breaking changes. Slightly less descriptive errors in extreme edge cases, but they're impossible to hit with our use of the dlopen library.

NullCharacter() => Not possible, since we control the one use of "NiFpga", and all other strings are function names which can't have null anyway.
NullSymbol => Can only happen when loading data symbols, which we don't do anyway.
AddrNotMatchingDll() => Somehow a pointer to a non loaded library is requested. Again not possible with how we expose.

I actually debated just panicing when any of these cases occured, but I think just a generic error is fine. The common errors are mapped properly to the FPGA status they would have normally had.