Refactor toir visit(StringExp) to avoid un-necessary computations.

[looked-at-me]

This doesn't include peekData() as described in #3490 not sure about the process about updating the interface files, if it's fine I'll just add it into this PR.

[kinke]

not sure about the process about updating the interface files

You could just add an inline method returning a DString, analogous to

ldc/dmd/expression.h

Lines 386 to 393 in 1494343

	#if IN_LLVM
	// The D version returns a slice.
	DString peekString() const
	{
	assert(sz == 1);
	return {len, static_cast<const char *>(string)};
	}
	#endif

[kinke]

Thx for figuring out the test failures. - Seems like the code duplication is pointless, so this should be further refactored. It doesn't have to be you, and it doesn't have to be in this PR, but it's ugly as hell once one knows about it.

[kinke]

Thx! I've pushed 2 more commits and am finally satisfied. ;)

[looked-at-me]

LGTM, the failure is due to a timeout. Not sure if you want to run it again?

[kinke]

I've learned to accept unfortunate sporadic failures and CI hickups. [Plus I can't retrigger or cancel Travis jobs as I used to.]

In case you're wondering about the formatting - we use clang-format for a consistent C++ style.