Optimize HTMLTree updates and deletions

[preciz]

• Accumulate deleted IDs to avoid O(N^2) List.delete overhead.
• Use Map.put instead of put_in for map updates.
• Use list comprehension instead of Map.take |> Map.values.

Performance improvements:

• ~100x speedup and ~200x less memory usage for filter_out in some benchmarks I ran with the big.html file
• ~1.3x speedup for find_and_update on large HTMLs