[nngeek195]

In previously server.py had many things to care, it had to handle web requests and  also at the same time it had to responsible for:

• Creating and destroying a process pool
• Counting how many diffs each worker had run
• Deciding when to recycle workers
• Recovering from crashed worker processes
• Deciding whether to shut the whole server down

As the issue #243 explains, the crash recovery logic was tangled inside the HTTP request handler. The pool state was also stored loosely in Tornado's settings dictionary instead of a proper separated object. This made it hard to understand and test independently from the web serving code.
So I extracted all process pool logic into a new class DiffExecutorManager in a new file executor.py.

executor.py now owns:

• Creating the ProcessPoolExecutor
• Tracking remaining_diffs to know when to recycle workers
• Retrying after a BrokenProcessPool crash
• Graceful vs immediate shutdown of worker processes
• Raising DiffPoolError when the situation is unrecoverable

server.py is now only responsible for:

• Receiving HTTP requests
• Fetching content from URLs
• Asking the executor manager to run a diff
• Formatting and returning the response

Because of this change, if the ProcessPoolExecutor crashes, the web server keeps running normally. Only the worker pool gets recycled inside DiffExecutorManager  and the HTTP server and request handling in server.py are completely safe and unaffected.

Close issue(#243)
@Mr0grog