Module caching initial concept issue

[damip]

Wasmer compilation takes time. We can improve things by saving the compiled modules in cache.

Let's make a finite-size cache of compiled Wasmer modules indexed by the bytecode hash:

• as Module objects when in RAM => they are ready to be instantiated
• in the future, if the RAM storage is not enough, also store as Serialized Modules on disk => they can then be deserialized, added to RAM cache, instantiated and executed

When a block containing uncached bytecodes needs to be exectued, threads should be launched to compile them all in parallel, put them in cache, and instanciate + execute them as they become ready.
When new bytecode is inserted/changed in the ledger, a separate thread should be launched to cache it as well, in order to be ready for its future execution.

Isn't hashing the bytecode too long ? => 127µs for 500kilobytes, 25µs for 100kilobytes
Improvement idea: in the bytecode field of the ledger, store the following:

• a varint u32 VM version/variant number
• the hash of the version + bytecode (to be updated every time the bytecode changes)
• the actual bytecode

That way, the module hash is readily available.

[AurelienFT]

Closing as it's the first version the second version and his design can be found in massalabs/massa#3488