fixed-size bignums

[Isaac Dupree]

data_structures/bignum.hpp

use bigint<> (signed) or biguint<> (unsigned) with a numeric template
parameter being the number of bits you want (it must be a multiple of
64). For example, bigint<192> is a 192-bit signed two's-complement integer.

All operations are defined-behavior and modulo (behind the scenes,
arithmetic all uses uint64_t because that's the simplest way to do it,
so we get this defined modulo behavior as a side effect).

* Moderately tested (tests/bignum_tests.cpp). Egregious bugs in
add/multiply/subtract squelched, but there may still be some un-found
incorrect results.

* No division operation yet. Has isqrt, ilog2, etc. already. Can
multiply from smaller to bigger int sizes using lossless_multiply() or
multiply_to<number of bits>().

* Works in a physical_quantity; doesn't work in bounds_checked_int yet.

* Has conversion to floating point but not currently from floating point.

* I might have left out something silly because I've only tested it in
test code yet, not real code.

Also: no arbitrary-precision compile-time numbers yet. I think they
will have to be semi-duplicate code with the runtime numbers; this
runtime bignum code has lots of loops and is definitely not constexpr.

-Isaac