bindnode: fix for stringjoin struct emission when first field is the empty string

[warpfork]

In the case of an empty string as the first field, the buffer length
is not a valid proxy for whether we're on the first field or not.

This means if we have some type like:
type Foo struct {a String; b String} representation stringjoin(":"),
and the value of it is {"", "b"}, then the string of that
should still be ":b".

Before this fix, it would incorrectly be emitted as "b" (no joiner),
which would not round-trip.

[rvagg]

you know you want to add a test for this

[mvdan]

Nice catch, oops.

FYI Rod, with Eric we've been discussing testing. One half of it will be extending the schema test suite, and another half will be adding table-driven bindnode tests for anything that's specific to how it handles Go types.

In this particular case, I agree with Rod: this could just be an added test in the schema test suite, so there's no reason to delay it.

[warpfork]

You're both right, this was easily added to the existing schema test suite, so I was indefensibly lazy not to add a regression test. Fixed.

(I yearn to get to data-driven tests that we can put in the ipld meta repo, but that's no excuse for not doing the best we can along the way until we get there.)