Add compile_tr_native() for Taptree-native policy compilation

[kwsantiago]

Closes #905

Adds compile_tr_native() which decomposes all Or and Thresh branches into separate TapTree leaves instead of using OP_IF/OP_NOTIF within a single leaf. Each leaf is compiled to branch-free Tapscript and assembled into a Huffman-weighted TapTree.

Also refactors enumerate_policy_tree into a shared enumerate_leaves helper to avoid duplication.

[apoelstra]

In ae5c7d3:

I like this approach. Letting the user provide a maximum and returning an error if we go past it is probably best for now. (Much) later I will overhaul the compiler API and provide a general set of control parameters, and hopefully we can unify all the different compilation functions into one, but for now I'd love to see this functionality:

Can you:

• Run cargo +nightly fmt to make the format job happy (if you don't have a nightly compiler let me know and I'll push a commit that does this)
• Remove the two dead doc links (see CI job) to MAX_COMPILATION_LEAVES and compile_tr (the latter maybe you just need to fully qualify)
• Add a unit test for an and-of-ors and(or(A,B),or(C,D)) and check that this explodes into four leaves and semantically matches the other compiler
• Add a unit test for a 3-of-5 threshold and check that it explodes into 10 leaves (unsure about checking semantic equivalence in this case; just lifting won't be sufficient; maybe just iterating through the leaves and asserting they're all different is enough?)

I'm a little suspicious of your threshold expanding logic. Before I step through it more carefully, I'd like you to add some unit tests just in case it's wrong.

Thanks!

[kwsantiago]

Ran nightly fmt, fixed the two doc links, and added tests for and(or(A,B),or(C,D)) (4 leaves + semantic match) and thresh(3,5) (10 distinct leaves). Also toned down the issue description re: your feedback.

[apoelstra]

Thanks! Clever use of entails() in a46031c -- I like it.