Mempool: streamline the implementation

• Remove dead / redundant code (e.g. valid operations are now stored and made available by the protocol, yet they are also still stored in the shell mempool...)
• The prevalidator_internal.ml and prevalidator_internal_common.ml files no longer make sense now that there is no legacy vs new mempool split anymore.
• Most of the mempool-related files are named prevalidator_..., but not all of them e.g. prevalidation.ml (which is a very uninformative name), shell_operation.ml (which is actually only used in the mempool), shell_plugin.ml, etc. There should be a consistent and meaningful naming scheme. Alternatively, can these files be moved to a separate lib?