I managed to further optimize the Blake3_u4 implementation (and fixed script size calculation).

• The new size is 68.7kb instead of 75.6kb (about 9.5% improvement).
• Stack consumption is slightly reduced, from 671 to 657.
• This optimization requires updating rust-bitcoin-script-stack with a new PR that introduces a script optimizer.
• One caveat: The stack needs to contain only the message before the script is called. Any additional elements must be moved to the alt-stack. That means an extra overhead of moving to and from the altstack, with that the result would still be under 70kb