Fix tuple_fetch with nonempty negative

[lukaszsamson]

recompute optional? instead of trusting projected value when tuple_take_element flags found?.

Note that it adds another reduce over negs that the short path if tuple_pair_projection_keeps_full_fst?(negative, bdd), do: value, else: ... tried to avoid

Fixes #15491

Assisted-by: Claude Opus 4.8 & GPT 5.5

[gldubc]

The diagnostic and the fix are correct, but i don't believe there is need for a new function tuple_line_index_absent?, and since i rewrote the entire function in #15668, I opted for a conditional check on the final line on tuple_bdd_fetch_static.