Tests: Do not assume empty vectors alias

[jakobnissen]

Currently, TS uses Base.mightalias to throw an error if the two inputs to three- arg transcode alias.
Two problems here: Base.mightalias is internal, and the new memory PR to Julia changes aliasing of zero-length arrays.
The temporary fix for now is to remove the test that tests that zero-length arrays trips the Base.mightalias. Later, let's fix this properly by not relying on Base internals in TS tests.

This fixes TranscodingStreams' test errors on JuliaLang/julia#51319

[vtjnash]

It is correct to assume that mightalias is the right function to use here. It is public. But just incorrect to assume that zero-element arrays have any aliasing as there is no memory region in common between them.

[jakobnissen]

How is it public? It's unexported, and it's docstring is not part of the Julia documentation. Nonetheless, it's good to hear that it can be used in packages :)

[vtjnash]

It is the public facing part of Base.dataids, which is documented as being intended for extension by custom arrays.

[nhz2]

Looks good to me.
Maybe there could be an additional check for input.data === output.data but that could be in a different PR, and I'm not sure it is needed.

[jakobnissen]

I'm just going to merge this now since it's a bugfix, to maintain compatibility with Julia's master branch.