Handling disconnects

Using the disconnect test showed the agent was silently closing on disconnect

gearman-go and original gearmand don't play well, possibly an issue?

1. It's easier to test with (it installs in the regular path), we need to start and stop it for network error tests
2. Can't hurt to test with both implementations!

Dodging the race condition with a sleep was fine, but we slept
for a whole second which was longer than this test takes to run!

I've scaled the sleep down to 250 miliseconds and increased the tries.

Even this is probably excessively long but now the test is fast agin.

Dodging the race condition with a sleep was fine, but we slept
for a whole second which was longer than this test takes to run!

I've scaled the sleep down to 250 miliseconds and increased the tries.

Even this is probably excessively long but now the test is fast agin.

Better Work() without Ready() behaviour, now with improved tests!

* FIX: committed test froze
* FIX: committed test had a race condition!
* Added properly handled panic test
* Timeouts so that these tests should fail now if something goes wrong instead of failing.

Now if you try to call Work() without calling Ready(), it will trigger an attempt to run Ready(), and will only panic if there is an error.

More helpful panic if you call Work() before you call Ready()

Make pool clients safe for concurrent access with mutex.

Printf fixes

SUBMIT_JOB_SCHED is deprecated, worker.Immediately should be removed.

Clarify timeout option in Readme

worker.Immediately has value 2 which means the job will timeout after 2 seconds (maybe not very intuitive?). To have no timeout one has to pass the value 0 which is also constant worker.Unlimited (more intuitive)

Read all data before decoding incoming packet

This fixes the case of a large packet being received, the current code stops processing as soon as >12 bytes are received,

Bring back ability to get function name

expose job handle/uniqueId