test: cover distribute_cascade and distribute_all_tokens at their exact limits

[IamHarrie-Labs]

Closes #333

What this covers

The existing tests only checked what happens when you go past the limits (distribute_cascade_exceeds_max_depth uses depth 6, distribute_all_tokens_too_many_tokens uses 11 tokens). Nothing exercised landing exactly on MAX_CASCADE_DEPTH (5) or MAX_DISTRIBUTE_TOKENS (10), which is exactly the boundary where an off-by-one (> vs >=) would hide.

distribute_cascade_at_max_depth: builds a chain six splits deep (parent plus five nested children down to a leaf account), so reaching the leaf genuinely requires cascading the full MAX_CASCADE_DEPTH. Calls distribute_cascade with max_depth set to the constant itself, asserts it succeeds, and checks every split in the chain ends at a zero balance with the leaf account receiving the full amount, not just that the call didn't error.

distribute_all_tokens_at_max_tokens: deposits exactly MAX_DISTRIBUTE_TOKENS (10) distinct tokens into a split and calls distribute_all_tokens with all ten. Asserts it succeeds, returns all ten distributions with the right amounts, and each token's balance is cleared afterward.

Both currently pass against the existing > check in lib.rs (not >=), so this documents the intended behavior at the boundary rather than fixing a live bug. If that comparison operator is ever flipped, these are the tests that would catch it.

Test plan

Ran all three required checks locally:

cargo fmt --all -- --check
cargo clippy --all-targets -- -D warnings
cargo test

All clean, 59/59 tests passing including the two new ones. Didn't touch anything under sdk/, app/, or indexer/, so no need to run the JS-side build for this change.