Async worker: Use Async tools to limit concurrency

[jlledom]

The asynchronous worker was using a custom method to control the amount of concurrent tasks in the reactor. However, Async provides some tools for this. Better use them rather than inventing the wheel.

Check: https://socketry.github.io/async/guides/best-practices/index

This also includes some minor changes to:

• Enable dotenv in tests
• Fix multithreading in worker async test

How to verify:

1. Set CONFIG_ASYNC_WORKER_MAX_CONCURRENT_JOBS to the desired number.
2. Use hey to generate fake load

   ./hey_linux_amd64 -z 60s -c 60 -q 20 "http://localhost:3001/transactions/authrep.xml?service_token=TOKEN&service_id=2&usage%5Bhits%5D=1&user_key=APP_USER_KEY"

3. Check with puts semaphore.count that CONFIG_ASYNC_WORKER_MAX_CONCURRENT_JOBS is never exceeded

[jlledom]

@akostadinov @mayorova I made some changes from your comments in the call:

• https://github.com/3scale/apisonator/pull/361/commits/f6c14e51c2c4a80200c39980c2f7b6be8f3633ac: Ensure stopping all tasks in case of exception.
• https://github.com/3scale/apisonator/pull/361/commits/4b257d8378e7be90972fbe0fcf612a6e88b2d252: Wrap entry points of the two threads under a Sync call which creates a new reactor, so it's clear which is the root reactor and where it's defined.
• https://github.com/3scale/apisonator/pull/361/commits/5c16cca51cf3d2b2ae88fa374ba27a6787f13493: Remove all calls to Async from the processor. This way they run sequentally because they are all in the same task under the parent reactor. All tasks under process_all and clear_queue are added to the same parent reactor, so the main task could get executed between tasks. This is prevented by the barrier, which makes the main task to wait until all children tasks finish.

[jlledom]

@akostadinov I implemented your suggestions