Split processors into monitor and user

[Christian-B]

Every single use of chip.processors was inefficient.

Typically the caller want just the ids and/ or just the user cores.

This pr adds extra methods to provide these more efficiently.

To add efficiency in the common case where just user processors (or their ids) are wanted the monitor and user processors are split into two dicts.

This comes at a slight cost for some methods using all processors but not enough to make is desirable to have duplicates dicts.
The only use cases that wanted all processors actually only wanted their ids so the new all_processor_ids method is faster anyway

---

Chip now also consistently uses machine_version().n_non_user_cores to determine the number of monitor processor and Chips just point to the global dict.
(A pointer to the global dict was required to allow some tests to hack in a different number of monitor cores)

tested by
http://apollo.cs.man.ac.uk:8080/blue/organizations/jenkins/Integration%20Tests/detail/chip_test/2/pipeline
(note that actually tested with Chip.processors completely remove which is why the different branch names)

Fixes #182

[coveralls]

coverage: 93.714% (+0.006%) from 93.708%
when pulling ee9d0ca on split_processors
into ecc3ae3 on master.