Remove DefaultCsprng

[jedisct1]

Reserve the name for the forthcoming global CSPRNG instead.

Not a lot of changes required since we don't use this type.

Add some comments by the way, as well as a Rand interface to getRandom().

[daurnimator]

I would keep this here and switch it over when the new one arrives.

[jedisct1]

The interface may be different, though.

[daurnimator]

Hmm, this makes me think std.crypto.randomBytes is badly location; perhaps should have its implementation moved here.

[jedisct1]

Having it in std.crypto also makes sense. Maybe we can reexport it.

[andrewrk]

I'm on board for some sweeping breaking changes to reorganize this, but let's leave this until we have the new thing, so that users have a clear upgrade path.

[jedisct1]

Sounds reasonable!

The removal of DefaultCsprng has been reverted, so the PR becomes adding a Rand interface to getRandom() and comments :)

[andrewrk]

proposal: delete this function, and follow this pattern:

zig/lib/std/heap.zig

Lines 88 to 98 in 1e13e8e

	pub const page_allocator = if (std.Target.current.isWasm())
	&wasm_page_allocator_state
	else if (std.Target.current.os.tag == .freestanding)
	root.os.heap.page_allocator
	else
	&page_allocator_state;
	var page_allocator_state = Allocator{
	.allocFn = PageAllocator.alloc,
	.resizeFn = PageAllocator.resize,
	};

Combined with moving it to std.crypto, usage would look like this:

std.crypto.os_random.bytes(buf);

Which I think would be enough to justify deleting std.crypto.randomBytes. This would be a good time to plan what the namespace of the #6704 API would be. Perhaps std.crypto.random meaning that we could tell users to upgrade from std.crypto.randomBytes to std.crypto.random.bytes which is a pretty satisfying upgrade, especially considering that bytes could then be replaced with any of the other Random interface functions, and it gains the benefits of the thread-local CSPRNG infrastructure.