Similar to several other targets, this commit override
`isLSRCostLess` to set instruction number as the first priority
when LSR pass deciding the cost.

Besides, this commit also takes the extra temporary register
may be used into account in `NumRegs`. This is same as riscv,
see the reason in https://github.com/llvm/llvm-project/pull/92296.

Copied from x86.

I'm considering a operator>(MCRegister, unsigned) and
operator<(MCRegister, unsigned) so I have not updated those lines. Such
comparisons are common on MCRegister.

ctype_utils/wctype_utils were chaged in
120689e46679c6db37cd9e839ec0721e80a22d4f and
e7f7973899f76773ae6e9a6b1e8c7e9f9cc5cb56, respectively to operate on
char/wchar_t. Now we can switch to the overloaded names (e.g. have noth
`isspace(char` and `isspace(wchar_t)`) to simplify the templatized
strtointeger implementation from
315dfe5865962d8a3d60e21d1fffce5214fe54ef and make it easier to
potentially add templatized strtofloat implementation.

Rename RegNum to Reg.

Close https://github.com/llvm/llvm-project/issues/166068

The cause of the problem is that we would import initializers and
pending implicit instantiations from other named module. This is very
bad and it may waste a lot of time.

And we didn't observe it as the weak symbols can live together and the
strong symbols would be removed by other mechanism. So we didn't observe
the bad behavior for a long time. But it indeeds waste compilation time.

NFCI -- the deleted copy constructor already made this immovable. The
explicit operations just make clear that this was intentional.