WriteAPIImpl.Flush is unnecessarily complex

[pabigot]

WriteAPIImpl.Flush() appears to have an irrelevant bug, and to be overly complex for what it actually does, which does not include
flushing as implied by the documentation.

waitForFlushing() has two steps:

1. Loop until the goroutine animating bufferProc() has acknowledged that bufferInfoCh has length zero. This is probably an error, and the length of bufferCh is intended to be checked.
2. Loop until the goroutine animating writeProc() has acknowledged that writeCh has length zero.

A fundamental problem is that bufferCh, bufferInfoCh, and writeCh are all unbuffered channels: their length will never be anything but zero. So the only effect of waitForFlushing() is transient synchronizations with the bufferProc() and writeProc() goroutines. pabigot@b9b12c8 provides a simplified equivalent.

Since cleaning this up doesn't address the core problem of #289 I haven't submitted a PR, but I can do so if you wish.