[Packer] Greedy Clustering Candidate Selector Class

[AlexandreSinger]

Outlined all of the logic related to selecting a candidate molecule to pack into a given cluster into its own class. This class will maintain the information required for computing gains and selecting the highest gain molecule to pack.

By separating out this logic, it makes it easier to modify and contribute to this code since it is now obvious where the code starts and where it ends.

I have also went through and tried to modernize anything I could (for example, replacing C-style arrays for vectors). This code still needs a lot more cleanup, but this PR will be a good stepping stone for future refactorings.

[AlexandreSinger]

QoR results for vtr_reg_qor_chain, run on my local machine:

	baseline_parse_results.txt	gr_can_sel_parse_results.txt
vtr_flow_elapsed_time	1	0.932162601
odin_synth_time
parmys_synth_time	1	0.933314613
abc_depth	1	1
abc_synth_time	1	0.936982841
num_clb	1	1
num_memories	1	1
num_mult	1	1
max_vpr_mem	1	1.000603156
num_pre_packed_blocks	1	1
num_post_packed_blocks	1	1
device_grid_tiles	1	1
pack_time	1	0.928684491
placed_wirelength_est	1	1
place_time	1	0.925694356
placed_CPD_est	1	1
min_chan_width	1	1
routed_wirelength	1	1
min_chan_width_route_time	1	0.933681889
crit_path_routed_wirelength	1	1
critical_path_delay	1	1
geomean_nonvirtual_intradomain_critical_path_delay	1	1
crit_path_route_time	1	0.950308163

It looks like my machine was loaded while running the baseline; however, what is important is that it looks like we get the exact same minimum channel width, CPD, and WL. This implies that my refactoring of the packer had no changes to the output clustering.

[vaughnbetz]

Looks good, thanks!
Some suggested commenting updates, and some functional updates that can be deferred until after the initial merge. Should definitely get rid of (or create an issue for) the non-deterministic random number sequence when attraction groups are used.