doc: add SFT length filter (15-token floor) with IFD bias rationale

Per-field length filtering in SFT drops instruction--response pairs
with responses shorter than 15 tokens. Short replies exhibit high
per-token variance in both conditional and unconditional loss
(Appendix A.3 / Figure 5), which would distort downstream IFD-based
difficulty estimates.

main.tex

@@ -154,8 +154,14 @@ pipeline proceeds as follows:
     previously kept sample $\mathbf{s}'$.
 \end{enumerate}
 
-An optional LLM-as-Judge scoring module provides multi-dimensional
-quality scores that can be used to filter low-quality samples.
+In addition to deduplication, instruction--response pairs whose
+response contains fewer than 15 tokens are discarded.
+Short replies exhibit high per-token variance in both
+conditional and unconditional loss (Appendix~\ref{sec:ifd_bias},
+Figure~\ref{fig:length_bias}), which would otherwise distort
+the IFD-based difficulty estimates used for downstream sample
+selection.  This length floor is applied per-field, analogous to
+the pretraining filter described above.
 
 \subsection{DPO Data Generation}