improve the std.fs.Dir API to not require an allocator

[andrewrk]

Extracted from a TODO comment:

zig/std/fs.zig

Lines 430 to 431 in bdfb314

	/// TODO remove the allocator requirement from this API
	pub fn open(allocator: *Allocator, dir_path: []const u8) OpenError!Dir {

Iterating over a directory does not fundamentally require an allocator. Let's remove the allocator requirement from this function.

This will remove one of the ways that this API can fail, making it more robust.

[ghost]

I've been trying my hand at this and I ended up frustrated each time I've tried a different approach. The challenge comes from Dir methods such as nextDarwin(), nextLinux() and nextBsd() where they all contain a side-effect of reallocating the buffer and thus doubling the len each time that side-effect happens. I don't know how it can get anymore robust than this without having it be not be dynamic without allocating and avoiding segfaults due to a lack of memory safety from the same allocation. I would love to solve this issue but I need some advice and direction, otherwise I feel lost and disappointed.