docs: apply three-tier drift ranking to pod deletion cost RFC (kp-flu)

Update the Pod Deletion Cost Controller RFC to use three-tier node
partitioning (drifted / normal / do-not-disrupt) instead of two-group
partitioning. Drifted nodes are prioritized for early draining so
scale-down events naturally assist drift progress.

Changes applied:
- Summary mentions drifted node priority
- Signal alignment section uses three groups (A/B/C)
- Ranking strategy describes drifted tier placement
- How-it-works steps and diagram updated to three tiers
- New example: drifted node drains first
- Appendix A RankingEngine description updated
- Appendix B do-not-disrupt example shows all three tiers

All references to the SLO drift budget RFC have been removed per
review feedback — the three-tier ranking stands on its own merits.

Author: nathangeology