Implement granular locking

[winitzki]

• When the reaction site decides on a new reaction, it should only lock the molecule values that it needs to examine. This will allow several reaction schedulers to operate concurrently, as long as they can operate on different molecule copies without contention.
• The locking should be both per-molecule and per-molecule value, such that a reaction could specify that it needs any 2 copies of a molecule regardless of their values, or it needs to examine potentially all copies of a molecule, etc.
• The reaction scheduler should hold a lock on a subset of molecules, on a subset of molecule values for a given molecule, and on a subset of copies of a molecule that has a given value.

Implement two locking principles:

1. Decreasing monotonicity: Once a lock is obtained on a subset of items, there cannot be a further lock obtained on a larger subset while still holding the lock on a smaller subset. The only change in locking can be the release of the lock on the entire subset or on a part of the subset.
2. Non-contention: We lock the items only because we cannot reach the correct decision without examining these items, and because other agents need to be prevented from withdrawing these items while we are still examining them. We release the lock on an item only when we learn that the correct decision can still be reached without using that item.

[winitzki]

Latest PRs increased granularity but did not implement the algorithms described here. It appears that these algorithms will not significantly improve performance or will degrade it, while introducing immense complexity into the imperative core of Chymyst.