Make the inner Jaccard loop cheaper

[hittjwX]

• See TidyObsidian/find-duplicate-blocks.py

For each candidate, you recompute jaccard(tokens, cb["tokens"]) by doing & and | on Python sets.
​
Change:

• Pre-store len(tokens) alongside tokens inside canonical_blocks, so you compute union size as len_a + len_b - inter and avoid building a full union set.

• If you keep tokens in a sorted list instead of a set, you can do an intersection with a two-pointer walk, which is often faster and more cache-friendly at this scale.

Both reduce per-comparison overhead without changing behavior.

[hittjwX]

Recent changes try to use a hash-based signature token, cache the token set size, and avoid building a full union set for Jaccard. TidyObsidian/find-duplicate-blocks.py