Support floating-point values other than 32 or 64 bits

[osandov]

long double is 80 bits on x86/x86-64. GCC also supports _Float128 on some platforms: https://gcc.gnu.org/onlinedocs/gcc/Floating-Types.html. GCC and Clang also seem to support _Float16.

Like for >64-bit integers (see #170), we support creating drgn.Type and reference or absent drgn.Object instances for non-32-/64-bit floats, but we don't support values. Unlike for integers, I don't know if there's a common arbitrary precision floating-point library we can use.