fix racy access of err variable

[jhump]

Without this fix, the handleSend function is closing over err from the outer function. This trips up the Go race detector because handleSend gets invoked in a goroutine and there is no synchronization between that goroutine and a subsequent write to err after a call to stream.Header.

This appears to be an accidental closure -- handleSend doesn't actually need to share the err variable so can instead just create its own err local variable. Which clears it up.

[jhump]

Ping @achew22. This is causing failing tests for some of our stuff because every bidi stream invocation tickles the race detector. We've temporarily forked grpc-gateway. This seems like a non-controversial fix, so we were hoping to not have to maintain a fork for very long.

[achew22]

Can you help me understand why changing the template doesn't change any of the generated code in the examples section? Could you run make examples and include any changes that are in there?

[jhump]

@achew22: I didn't even realize that there was an examples folder.

I ran make generate, but it produced a ton of differences. I am guessing I have a different version of swagger-codegen than was used to produce the existing files. (I actually just installed it on OS X: brew install swagger-codegen. So maybe the existing swagger-generated files need to be freshened?)

I only git add'ed the ones relevant to my change.

[jhump]

ping: @achew22, I think this branch is ready to go.

[achew22]

Could you rebase this change? I've fixed the test that was not running that I was expecting to fail in here.

Thanks!

[jhump]

Could you rebase this change?

@achew, done