Optimize aligned memory used by the runtime

[jackcmay]

Problem

The AlignedMemory implementation added a new and superfluous memset in cases where the data was later written incrementally (most pronounced in program parameter serialization). This is causing performance degradation.

Summary of Changes

• Rework the aligned memory implementation to remove the memset, includes other optimizations utilized in other areas of the rbpf (and thus runtime)
• Added serialization benches

Fixes #

[jeffwashington]

raw results on master. serialize 78s with this change vs 108s without this change

jack:
ledger processing timing ExecuteTimings check_us 1623487 load_us 30607593 execute_us 257297503 store_us 70930243 details ExecuteDetailsTimings serialize_us 78524207 create_vm_us 12694585 execute_us 39198653 deserialize_us 41360025 changed_account_count 212316 total_account_count 5313048 total_data_size 341233366608 data_size_changed 4066129474
ledger processed in 3 minutes and 24 seconds. root slot is 76820740 1 fork at 76820780 with 37 frozen banks

master
ledger processing timing ExecuteTimings check_us 1546815 load_us 28623897 execute_us 287027755 store_us 65587266 details ExecuteDetailsTimings serialize_us 108053806 create_vm_us 12027731 execute_us 39847740 deserialize_us 40752505 changed_account_count 212316 total_account_count 5313048 total_data_size 341233366608 data_size_changed 4066129474
ledger processed in 3 minutes and 36 seconds. root slot is 76820740 1 fork at 76820780 with 37 frozen banks

[jeffwashington]

1.6 looks good at first glance. I will work this into the graph. I ran jack's change twice to make sure it really was that fast:

jack 1.6:
ledger processing timing ExecuteTimings check_us 1813310 load_us 163932199 execute_us 283908612 store_us 127102701 details ExecuteDetailsTimings serialize_us  41239102 create_vm_us 12642299 execute_us 40188027 deserialize_us 35009657 changed_account_count 418476 total_account_count 5834166 total_data_size 372337745651 data_size_changed 16219497930
ledger processed in 5 minutes and 1 second. 55019 MB allocated. root slot is 76820740 1 fork at 76820780 with 37 frozen banks
ledger processing timing ExecuteTimings check_us 1785740 load_us 163823277 execute_us 283017909 store_us 127010841 details ExecuteDetailsTimings serialize_us 41133678 create_vm_us 12580657 execute_us 39946844 deserialize_us 35045918 changed_account_count 418476 total_account_count 5834166 total_data_size 372337745651 data_size_changed 16219497930
ledger processed in 5 minutes and 1 second. 55067 MB allocated. root slot is 76820740 1 fork at 76820780 with 37 frozen banks

1.6:
ledger processing timing ExecuteTimings check_us 1829167 load_us 167772692 execute_us 406036035 store_us 127593625 details ExecuteDetailsTimings serialize_us 153063242 create_vm_us 14000456 execute_us 47955506 deserialize_us 28170237 changed_account_count 418476 total_account_count 5834166 total_data_size 372337745651 data_size_changed 16219497930
ledger processed in 6 minutes and 5 seconds. 57645 MB allocated. root slot is 76820740 1 fork at 76820780 with 37 frozen banks

[jeffwashington]

confirmed that #17324, cherry-picked to v1.6, restores performance of replay. metrics spreadsheet for 1.6 updated

[jeffwashington]

updated metrics spreadsheet for master. looks good