Tested by transferring 29GiB data, around 246569 files
over 100Gig network.
- master branch
```
time mc mirror /root myminio/testbucket/
real 2m10s
```
- after changes
```
time mc mirror /root myminio/testbucket/
real 1m2s
```
By default runtime.NumCPU() workers are started.
Each 5 seconds, calculate current bandwidth and add new workers as
bandwidth gets increased. Stop adding workers if it is not useful
for increasing bandwidth.
This PR removes --parallel flag and implements a dynamic way of adding
threads to accelerate mirror/cp command.
After the calculation of objects that need to be copied/removed, mc
sends tasks (upload/remove) to a parallel manager which monitors the
progress of tasks and decide to allocate more threads to it to
accelerate the overall work.
This is very useful when we copy/mirror many relatively small files.
