[PERFORMANCE] Use `ConcurrentDictionary` for `Single`/`Ordered`/`Critical`

[computablee]

Identify the function(s) which are causing performance issues.

The current implementations of Parallel.Single/Ordered/Critical are locking a dictionary before modifying it. We can use a similar approach from the tasking system and use ConcurrentDictionary classes to implement lock-free modifications.

Describe the performance expected vs. performance observed.

Performance metrics still need to be measured. A comprehensive evaluation should be provided when a corresponding PR is opened.