set hash bins to 65k

[jeffwashington]

Problem

more bins breaks the accounts for scanning into smaller groups, allowing higher paralleism.

Summary of Changes

65k is our current structural limit and is probably a good stopping point for the moment.
Fixes #

[codecov]

Codecov Report

Merging #17912 (5aaba21) into master (361c1bd) will increase coverage by 0.0%.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master   #17912   +/-   ##
=======================================
  Coverage    82.6%    82.6%           
=======================================
  Files         431      431           
  Lines      121184   121179    -5     
=======================================
+ Hits       100158   100174   +16     
+ Misses      21026    21005   -21     

[jeffwashington]

colo, mnb snapshot.
Total improves slightly with the current algorithm. The algorithm is about to improve. Data below.

65k:
calculate_accounts_hash_without_index accounts_scan=1189663i eliminate_zeros=217116i hash=71724i hash_time_pre_us=71509i sort=202693i hash_total=70524929i flatten=1283538i storage_sort_us=1393i unreduced_entries=71426722i collect_snapshots_us=51704i num_snapshot_storage=345908i total=3017831i
calculate_accounts_hash_without_index accounts_scan=1073849i eliminate_zeros=198102i hash=67582i hash_time_pre_us=67410i sort=762645i hash_total=70524929i flatten=1155190i storage_sort_us=1532i unreduced_entries=71426722i collect_snapshots_us=72095i num_snapshot_storage=345908i total=3330995i
normal:

[jeffwashington]

some raw data:
upshot:
higher bins + better algorithm makes better performance:
lemond, 147M accounts, 200k slot subset of kin simulation accounts

current algorithm
us
9169758 - 256 bins
7081799 - 65k bins

refined algorithm with no flatten
6399554 - 256 bins
5093599 - 65k bins