It's not clear to me that it improves the generated code much, but
perhaps I'm simply not skilled in reading CCL disassembly.

This is more consistent with what Common Lisp already
does (e.g. READ-SEQUENCE), as well as being slightly more efficient.

Use dummy constructors for the decoding bits, since we're not using the
new scheme there yet.  base85 tests do not yet pass.

It's faster to check than '(UNSIGNED-BYTE 8) and more consistent with
the other return values.