use a one liner to clone the object where possible

- changed _treeMetadataMap to use the unproxified version of the tree, instead of proxified
- merged _parenthoodMap into _treeMetadataMap
- added a new symbol _metadataSymbol used to access the metadata from the proxified object, but only if you have a reference to the symbol

skip generation of a replace operation, if an object within the tree is replaced with a proxified version of itself by external code

Warning, this commit changes the performance of the benchmark. Generation of tested object is now 2x faster. This is good for the benchmark, because cheaper object generation allows the benchmark to be more sensitive in catching the differences in the performance of JSONPatcherProxy.

Just remember that the results from before and after this commit are not comparable.

Benchmark results before this commit
=============================
jsonpatcherproxy generate operation
 Ops/sec: 132404 ±2.55% Ran 8276 times in 0.063 seconds.
jsonpatch generate operation
 Ops/sec: 106585 ±6.73% Ran 6434 times in 0.060 seconds.
jsonpatcherproxy mutation - huge object
 Ops/sec: 906 ±2.45% Ran 51 times in 0.056 seconds.
jsonpatch mutation - huge object
 Ops/sec: 862 ±1.52% Ran 47 times in 0.055 seconds.
jsonpatcherproxy generate operation - huge object
 Ops/sec: 811 ±1.99% Ran 46 times in 0.057 seconds.
jsonpatch generate operation - huge object
 Ops/sec: 847 ±1.20% Ran 47 times in 0.055 seconds.
PROXIFY big object
 Ops/sec: 1813 ±2.00% Ran 102 times in 0.056 seconds.
DIRTY-OBSERVE big object
 Ops/sec: 1625 ±1.48% Ran 91 times in 0.056 seconds.

Benchmark results after this commit
=============================
jsonpatcherproxy generate operation
 Ops/sec: 136915 ±2.46% Ran 8566 times in 0.063 seconds.
jsonpatch generate operation
 Ops/sec: 104113 ±7.19% Ran 6330 times in 0.061 seconds.
jsonpatcherproxy mutation - huge object
 Ops/sec: 1960 ±3.53% Ran 130 times in 0.066 seconds.
jsonpatch mutation - huge object
 Ops/sec: 1607 ±1.84% Ran 88 times in 0.055 seconds.
jsonpatcherproxy generate operation - huge object
 Ops/sec: 1562 ±3.40% Ran 101 times in 0.065 seconds.
jsonpatch generate operation - huge object
 Ops/sec: 1620 ±2.04% Ran 90 times in 0.056 seconds.
PROXIFY big object
 Ops/sec: 3968 ±4.07% Ran 294 times in 0.074 seconds.
DIRTY-OBSERVE big object
 Ops/sec: 3285 ±1.83% Ran 180 times in 0.055 seconds.

the benchmark methods are almost exactly the same, but I have made the following changes:
- group benchmarks by the topic
- give more meaningful titles
- wherever the preparation code is not meaningful for the benchmark, move it outside the benchmark
- use code blocks to give scope to preparation variables

this is a change that makes the benchmark results incomparable with the previous versions.

however, it will work if you copy this file onto an older version and run it.

as requested in #39 (comment)

the benchmark simply serializes the whole object, which triggers the "get" trap on every subtree

by default such comparison will be excluded. Benefits of this:
- faster execution
- shorter report to copy to PRs

to make functionality equal to fast-json-patch'es