ended earlier.

add bin to package.json

failure output format for `pm uninstall`. A related reliability issue
with readAll() was found as well, that caused a timeout instead of a
proper rejection. Fixes openstf/stf#306.

disabled but as it turns out I simply misunderstood the documentation.
Fixes #41.

Further improvements to TCP/USB reliability. It was possible to get an error if the client socket (`adb connect` side) died first.

Make sure tests always use the CoffeeScript source, as they'll otherwise mismatch class instance checks among other things.

Pass error from connection to server. Usually happens if the device can't be found. May occur more than once.

Swap localId and remoteId so that they're not super confusing. Previously 'local' meant 'their local' and 'remote' meant 'our local'. This was caused by reading the adb source too closely and not swapping the names originally.

Almost entirely rewrite the TCP->USB adapter. Unfortunately there are still problems when a service responds with a failure status.

Various reliability improvements to stop the tcp to usb bridge from disconnecting in the middle of everything. The exact issue remains unknown, as it's not reliably reproducible. It seems, though, that this patch does the trick for now.

Dump everything adbkit writes to the server if the ADBKIT_DUMP environment variable is set. Output goes to `adbkit.dump` in the current directory.

Make sure we try to pump immediately in case readable already fired. Unfortunately this does not fix the disconnect problem even though very suspect.