Read BDT score from first 12 bits of xTOB

The BDT score is read from the first 11 bits of word 0 of the xTOB while it should be read from the first 12 bits instead, since the padding MSB bits are all 0 if the BDT score takes less than 12 bits. For now there were no issues with this since the BDT score takes only 11 bits, but the purpose of this fix is to avoid issues in the future if the BDT is re-trained and would result in a 12-bit BDT score, in which case the MSB bit would have been ignored.

See https://indico.cern.ch/event/1354773/contributions/5704602/attachments/2769314/4824688/2023-12-11.pdf, slide 6

This was tested by running a derivation with and without this change and verifying that BDT score histograms are identical.