Library Forwarding: Don't map float/double to fixed-size integers

[neobrain]

d04c94f unintentionally changed behavior on 64-bit applications: In particular, the apitraces attached to #3488 (comment) show that the LOD bias passed to glTexParameterf differs before and after the change:

< 12916 glTexParameterf(target = GL_TEXTURE_2D, pname = GL_TEXTURE_LOD_BIAS, param = 0)
> 12916 glTexParameterf(target = GL_TEXTURE_2D, pname = GL_TEXTURE_LOD_BIAS, param = -4184224)

It turns out GetTypeNameWithFixedSizeIntegers erroneously mapped float/double types to fixed-size integers, which led to wrong semantics (presumably bitcasting followed by rounding conversion, instead of plain copying). This mapping was done consistently for both host and guest, so this issue escaped my initial attempt at finding differences.

By fixing GetTypeNameWithFixedSizeIntegers, thunkgen now emits GuestWrapperForHostFunction<void(uint32_t, uint32_t, float), uint32_t, uint32_t, float>::Call for glTexParameterf, where previously the floats would have been replaced by uint32_t. Similar type checks have been updated to use builtin && integer for consistency without functional change.

Fixes #3455.
Supersedes #3488.