parallelize mpt updates

[kunxian-xia]

To turn on the parallel mode of witness assignment, set the env var PARALLEL_SYN=true. For example, you can run unit test tests::multiple_updates using

PARALLEL_SYN=true RUST_LOG=debug cargo test --release --lib tests::multiple_updates -- --nocapture

TODO list

• batching inversions in difference_is_zero gadget.
• parallelize canonical_representation.
• parallelize mpt padding rows (0.4s).
• parallelize mpt table assignment (1s). (fixed by scroll-tech/zkevm-circuits@a6f2288)
• parallelize the key_bit_lookups function (1s).
• add benches in which there are at least 10k mpt updates and get real world performance data.

[kunxian-xia]

This is related to #94 #95 #96.

[z2trillion]

Looks good. for future reference, when we run this on the mpt updates from chunk_83587.zip, the assignment time should drop from ~40s to 1.5s.