Region tree renaming and fix

[sergeypospelov]

This PR fixes issues with key deduplication by RegionTree and simplification found by @Saloed. Also, made renaming in RegionTree to make it easier to understand.

Details

Issue with key deduplication

Consider the following example:

// writings
arr {i <- 1} {j <- 2} {j <- 3}
// expected
arr {i <- 1} {j <- 3}
// actual
arr {j <- 3}

It happens, because in case keyFilter returns true, localize skips not only the top node, but also its subtree, which is incorrect in such case. Now localize with filtering maintains this correctly, but may leave some nodes even they satisfy keyFilter, which is OK for us.

Issue with simplification

Consider the following example:

reading := arr {i <- 1 | p1 } {i <- 2 | p2} [j]
model := {i -> 0, p1 -> true, p2 -> false, j -> 0}
composed := compose(reading, model)
// expected
1
// actual
RE in composition

It happens, because guarded update nodes still remains in the memory region, though guards of some of them have evaluated to false. Fixed it in memory updates map functions.

Renaming

RegionTree is actually a persistent event-storage with support of grouped reading and writing, where events live in specific regions. Thus, now its signature looks as following class RegionTree<Reg, Value>. IMO, it's much clearer.