Test if seekstart after write deletes data.

[nhz2]

This PR adds tests showing that currently calling seekstart on a stream in write mode can silently delete data.

[mkitti]

So both the tests here are broken? Is the plan to unbreak them?

[nhz2]

Yes. For the QuadrupleCodec case I think the fix is to have the fallback TranscodingStream seekstart throw an ArgumentError if the stream is not in read or idle mode. For the NoopCodec case, I think the solution is a little trickier because it doesn't use the fallback seekstart.

[mkitti]

Should we merge this then or are we waiting for something?