Broken benchmarks

[datapythonista]

The benchmarks are currently broken, and are making red the CI of all PRs that modify something in asv_bench (we run the CI check for benchmarks when something is modified in that directory).

This has probably been caused by #26225. But it's strange that the CI for that PR wasn't red, since it run the benchmark checks.

This is blocking #26414, #26703. The error can be seen here: https://dev.azure.com/pandas-dev/pandas/_build/results?buildId=12407

[ 16.10%] ··· ...ervalIndexMethod.time_intersection_duplicate         1/3 failed
[ 16.10%] ··· ========== ==========
                param1             
              ---------- ----------
                 1000     1.76±0ms 
                100000    73.3±0ms 
               10000000    failed  
              ========== ==========

[TomAugspurger]

Taking a look now.

[qwhelan]

It appears this large N never worked. I have a fix that yields a large speedup (~10x+) but possibly changes the intent of this benchmark: #26711