Effect of replacing RMW atomics with load non atomic modify store in OL critical sections

The initial implementation used relaxed_atomic in optimistic lock critical sections, including writer sections. This resulted in increments and decrements compiling to atomic RMW operations, where a non-atomic equivalent would suffice. Fixed by along the lines of replacing value.fetch_add(1, std::memory_order_relaxed) with value.store(value.load(std::memory_order_relaxed) + 1, std::memory_order_relaxed).

The effect of this change:

$ python3 ../../3rd_party/benchmark/tools/compare.py benchmarks ../../../unodb/build.release/benchmark/micro_benchmark_key_prefix ./micro_benchmark_key_prefix --benchmark_filter=olc_db --benchmark_repetitions=9
...
unpredictable_get_shared_length<unodb::olc_db>_pvalue                     0.0004          0.0004      U Test, Repetitions: 9 vs 9
unpredictable_get_shared_length<unodb::olc_db>_mean                      -0.0101         -0.0086             2             2             2             2
unpredictable_leaf_key_prefix_split<unodb::olc_db>_pvalue                 0.0004          0.0004      U Test, Repetitions: 9 vs 9
unpredictable_leaf_key_prefix_split<unodb::olc_db>_mean                  -0.0373         -0.0374            52            50            52            50
unpredictable_cut_key_prefix<unodb::olc_db>_pvalue                        0.0004          0.0004      U Test, Repetitions: 9 vs 9
unpredictable_cut_key_prefix<unodb::olc_db>_mean                         +0.0079         +0.0075            51            52            51            52
unpredictable_prepend_key_prefix<unodb::olc_db>_pvalue                    0.0004          0.0004      U Test, Repetitions: 9 vs 9
unpredictable_prepend_key_prefix<unodb::olc_db>_mean                     -0.0044         -0.0047            63            63            63            63