Parallel hash computations

[and-cb]

This PR makes hash computations happen in parallel in a thread pool. While hashes are computed in the background, a temporary, zero-byte hash is written to the pages. The zero hashes are replaced with the actual hashes once all changes have been processed.

Status of this PR: All tests are passing, but this is a very early draft that needs a lot of cleanup and refactoring.

TODO

• Put the DeferredChanges struct inside Context.
• Add helper methods so that the pair slotted_page.set_value(...); deferred.set_value(...) becomes 1 call.
• Currently DeferredRlpNode has replaced RlpNode almost everywhere. In practice, DeferredRlpNode is only needed when writing values, but when reading, RlpNode is enough, and has a lower overhead. Pointer should become a generic Pointer<R> so that both RlpNode and DeferredRlpNode can be used as appropriate.'
• Remove unnecessary premature computation of hashes.
• Currently, once a hash is computed, the whole node is rewritten, instead of just the hash. This can be improved.
• Remove lazy_static.
• Remove debug statements.
• Make the thread pool configurable (e.g. setting the number of threads)

[cb-heimdall]

🟡 Heimdall Review Status

Requirement	Status	More Info
Reviews	🟡 0/1	Denominator calculation Show calculation 1 if user is bot 0 1 if user is external 0 2 if repo is sensitive 0 From .codeflow.yml 1 Additional review requirements Show calculation Max 0 0 From CODEOWNERS 0 Global minimum 0 Max 1 1 1 if commit is unverified 1 Sum 2