Fix swapped performance metrics in RetBleed benchmark table

[2045castor]

This PR corrects a data entry error in the benchmark results table where the raw performance metrics (MEAN, MEDIAN, STDDEV) for "retbleed=off" and "retbleed=ibpb" were accidentally swapped.

The original table illogically showed the heaviest mitigation (ibpb) yielding the highest performance, while the unmitigated baseline (off) had the lowest. The correctness of this fix is verified by the existing Overhead column: calculating a 57.74 percent overhead from a true baseline of 91385.800 mathematically yields the 38619.000 result.

This commit simply maps the raw data back to the correct rows to align with the calculated overheads and expected hardware behavior.

[2045castor]

@matrizzo Since you added the original documentation, could you take a quick look?

While studying this repository, I noticed that the performance data (MEAN, MEDIAN, STDDEV) for retbleed=off and retbleed=ibpb appear to be accidentally swapped in the benchmark table. The calculated Overhead percentages strongly support this.

Could you please review and confirm if this is the case? Thanks!

[matrizzo]

Hi! It looks like the rows got mixed up somewhere, nice catch!

[2045castor]

Hi! It looks like the rows got mixed up somewhere, nice catch!

Thanks for the quick merge!

Since we're already fixing swapped metrics today, any chance we could accidentally swap the open and merged status for PR #380 too? 👀

(Jokes aside, would love a final look on that one whenever you have a moment!)