go/scanner: inconsistent handling of NUL bytes in 1.17

[findleyr]

Discovered via fuzzing using the dev.fuzz branch: CL 308611 optimized the scanning of identifiers, but inadvertently changed the handling of zero-bytes.

For example, when scanning the following bad input, go 1.17 is missing an error "illegal character NUL", though it does scan the same token stream (including a follow-up error for "illegal character U+0000"): "package p;e\x00"

This might be an improvement, but we should not unintentionally change this behavior for 1.17. The fix is trivial.

CC @griesemer

[gopherbot]

Change https://golang.org/cl/329790 mentions this issue: go/scanner: fall back to next() when encountering 0 bytes in parseIdentifier