Should we make TrieBuilder off-heap

### Description

`TrieBuilder` has replaced FST in the block term index. I'm wondering if it makes sense to also make it off-heap, enabling use case such as building a large term trie (similar to `FSTPostingFormat`).

Start by reading TrieBuilder's role in the block term index and compare it with FSTPostingFormat, which the issue cites as a similar use case. Determine whether off-heap construction is technically and operationally justified for large term tries; done means a documented design decision and an agreed implementation scope.

Written by the indexing model from the issue text.