perf: use vec_map paging mmu for executor memory

[tqn]

Some areas to potentially improve:

• memory_checkpoint always allocates as much space (for page pointers) as the largest key (page address). This is not necessary. Perhaps it should be a VecDeque of Pages along with its least address, or perhaps a HashMap or BTreeMap would do better. This choice probably depends on the shard size and memory access patterns.
• memory_checkpoint is PagedMemory<Option<MemoryRecord>> which is internally stored as Vec<Option<Option<MemoryRecord>>>. This is not the best for space.
• Loads of constant arithmetic at the top of memory.rs. Could be made into const functions taking (the size of) the generic value V.

[github-actions]

SP1 Performance Test Results

Branch: tqn/perf-executor-slowdown
Commit: b2aaa89b
Author: tqn

program	cycles	execute (mHz)	core (kHZ)	compress (KHz)	time	success
fibonacci	11291	0.02	5.79	0.15	1m17s	✅
ssz-withdrawals	2757356	3.42	59.12	21.99	2m53s	✅
tendermint	12593597	4.69	151.53	93.81	3m40s	✅