TST: Avoid floating point errors in asinh ticker

[greglucas]

PR Summary

This avoids the floating point issues with tests in the asinh tick locator by moving away from the 100 log10 cliff. This was failing with the new AVX instruction updates in numpy. The underlying issue here is that the floating point accuracy around 100 changes from one side to the other and the floor() call doesn't accept precision arguments. This is not what we are trying to test, so just move the test values away from that edge.

Closes #22708

Showing the tests passing on my fork: https://github.com/greglucas/matplotlib/actions/runs/2177190522

PR Checklist

Tests and Styling

• Has pytest style unit tests (and pytest passes).
• [N/A] Is Flake 8 compliant (install flake8-docstrings and run flake8 --docstring-convention=all).

Documentation

• [N/A] New features are documented, with examples if plot related.
• [N/A] New features have an entry in doc/users/next_whats_new/ (follow instructions in README.rst there).
• [N/A] API changes documented in doc/api/next_api_changes/ (follow instructions in README.rst there).
• [N/A] Documentation is sphinx and numpydoc compliant (the docs should build without error).