Remove duplicate 99% KLD output, add additional percentiles to match mainline

[AesSedai]

There was a PR recently on mainline to add the additional percentiles to llama-perplexity here and since I was doing some perplexity testing with ik_llama, I figured I'd put up a PR to fix the 90% being listed twice, and add in the other percentiles to match mainline.

I compiled and ran a llama-perplexity run to verify that the output is showing correctly:

...
====== KL divergence statistics ======
Mean    KLD:   0.017475 ±   0.000763
Maximum KLD:  11.152802
99.9%   KLD:   1.098822
99.0%   KLD:   0.221226
95.0%   KLD:   0.055542
90.0%   KLD:   0.029520
Median  KLD:   0.004062
10.0%   KLD:   0.000020
 5.0%   KLD:   0.000005
 1.0%   KLD:   0.000000
 0.1%   KLD:  -0.000003
Minimum KLD:  -0.000423
...

• I have read the contributing guidelines
• Self-reported review complexity:
  • Low
  • Medium
  • High