Closes #1300 - Improve Performance of DataFrame Display

[Ethan-DeBandi99]

Closes #1300

• Updates the _get_head_tail() method to function by creating a single server message instead of 1 per column in the DataFrame. This was done by adding _get_head_tail_server() which allows us to benchmark the new code against the old. The old function has been retained for benchmarking.
• Adds a DataFrameIndexingMsg.chpl to parse and process the server message for indexing the columns. The message is configured so the the column type, column name (in the df), column object name (server-side) are sent to the server to indexing to the appropriate head/tail.
• Benchmarking configured to test _rep_html() which now calls _get_head_tail_server(). The benchmark also checks _get_head_tail_server() and _get_head_tail() directly to allow for easy comparison.

Per the issue description, @joshmarshall1 and I did this without using aggregation because the datasets should be relatively small.

Benchmarking on a single node shows an performance already, but we will need to benchmark this on a multi-node system. Results from single node:

array size = 10,000
number of trials =  5
>>> arkouda dataframe display
numLocales = 1, N = 10,000
  _repr_html_ Average time = 0.0401 sec
  _repr_html_ Average rate = 0.03 GiB/sec
  _get_head_tail_server Average time = 0.0406 sec
  _get_head_tail_server Average rate = 0.03 GiB/sec
  _get_head_tail Average time = 0.0825 sec
  _get_head_tail Average rate = 0.01 GiB/sec

@reuster986 - I will leave it up to you if you would like to request an out of band benchmark.

[Ethan-DeBandi99]

Added @joshmarshall1 as a reviewer even though he helped write this code. There are some elements that I handled that would be good for him to review as well.

[joshmarshall1]

As a participant in developing this, I won't approve it, but after a second review I think we should go through and add comments throughout the chpl code, since there are almost none, just to make future maintenance easier. I also noticed a few places that TODO tags were left in and a stray #

[reuster986]

Looks good overall! Just a couple requested changes in the benchmark, and a suggestion about when/how we switch to the new implementation.

[Ethan-DeBandi99]

Updated benchmark results on single node with requested updates from @reuster986 implemented.

array size = 10,000
number of trials =  5
>>> arkouda dataframe display
numLocales = 1, N = 10,000
  _get_head_tail_server Average time = 0.0261 sec
  _get_head_tail_server Average rate = 0.05 GiB/sec
  _get_head_tail Average time = 0.0547 sec
  _get_head_tail Average rate = 0.02 GiB/sec

[reuster986]

Looks great, thanks!

[stress-tess]

Couple comments, nothing major. The only thing that might need to be updated is "support uint for segarray values" question? But the logic all looks good to me

[stress-tess]

Looks good!