BUG: irr function inconsistent behavior on different cpu architecture

Describe the issue:

Hi, I noticed that npf.irr function could return different results on different cpus in some cases. The results would match if i run irr function more than once. Please see example below.

Reproduce the code example:

python -c "import numpy_financial as npf; amount = 37155; month_count = 144; monthly_payment = 560; npf.irr([-amount] + [monthly_payment] * int(month_count));  print(npf.irr([-amount] + [monthly_payment] * int(month_count)))"

python -c "import numpy_financial as npf; amount = 37155; month_count = 144; monthly_payment = 560; print(npf.irr([-amount] + [monthly_payment] * int(month_count)))"

Error message:

On aws ec2 6i instance with Intel Ice Lake, I got the following:
0.01258179635558565
0.01258179635558565

On aws ec2 7i instance with Intel Sapphire Rapids, I got the following:
0.01258179635558565
0.006850691458678115

Runtime information:

1.23.5
3.10.14 | packaged by conda-forge | (main, Mar 20 2024, 12:45:18) [GCC 12.3.0]

Context for the issue:

No response

Start by running the two provided npf.irr command lines on the reported CPU architectures and inspect the irr entry point. The fix is complete when repeated calls produce the same result across architectures, with coverage for the reproduced case.