mesa: don't leak lanes

[dozreg-toplud]

Resolves #

[joemfb]

The prior code was certainly very silly, but this change should be a no-op wrt refounting. u3do() transfers the sample, so gaining while calling it should be no different from just dereferencing the tail. After the line in question, both las and dat should still be live.

[dozreg-toplud]

the difference is that in prior code the reference to the tail is counted and subsequently leaked, and now it is an uncounted reference

[dozreg-toplud]

so now u3z(dat) also frees las, and previously it didnt, there was an observable leak

[joemfb]

Right, thanks. I was focused on dat and didn't notice that the old behavior was effectively las = u3k(u3t(dat));.