This commit re-implements the benchmark tool, doing a lot of refactoring and
speeding up things, while reducing memory usage. Some notes:
1. The `testkey` subcommand was removed. This was needed because it makes no
sense to have it with data types like lists. I'll leave support for creating
test data for another PR (if at all needed)
2. The bench tool now uses lesser memory "in a go." Previously, all the get,
set and update packets were generated in one single step. This is no longer
done. Instead, we generate the packets for one type, run its corresponding
tests and completely deallocate and cleanup once we're done with that test.
This helps alleviate memory usage.
3. Instead of going in the sequence (GET, SET, UPDATE, ...) and running that
set some n times, we now run each test n times, and repeat that for the other
tests. The advantage of this is code clarity.
4. Finally, more validation checks were added ensuring that the benchmark tool
very closely resembles a "real-life" scenario wherein a connection is picked up
from a pool by a thread, a task run-on and evaluated.
5. Better logging and error reporting: instead of vaguely crashing, the
benchmark tool provides more clarity on errors. However, panics in worker
threads are yet to be improved in terms of reporting.
It's important that we report an error when the sample space is too
low for the given k/v size and key count. Previously, we'd end up
entering an infinite wait time (essentially an infinite loop); this
has been fixed.
The create table query was being run with the workpool that also
attempts to switch to it on init (pre-loop stage). This caused us to
read in lesser bytes (since we use read_exact) and after the connection
is closed, the server would respond with a broken pipe error as it fails
to write data to the closed connection (since it expected to write more)