Axes.axes.secondary_xaxis does not produce good ticks with some transformations

Nobody has claimed this yet.

### Bug report

**Bug summary**

If I supply a nonlinear transformation to `secondary_xaxis`, it produces overlapping tick positions, at least when I do not use a logarithmic scale, see images below.

**Code for reproduction**

```python
import matplotlib.pyplot as plt
import numpy as np

c = 299792458.0

fig, ax1 = plt.subplots()
ax1.plot((1e16, 1e19), (0, 1))
ax1.set_xlim(1e16, 1e19)
ax1.set_xlabel(r'$\omega/\mathrm{s^{-1}}$')
ax2 = ax1.secondary_xaxis(-0.15, functions=(
lambda omega: 2 * np.pi * c / omega,
lambda lambda_: 2 * np.pi * c / lambda_)
)
ax2.set_xlabel(r'$\lambda/\mathrm{m}$')
plt.tight_layout()
plt.savefig('issue1.png')

ax1.set_xscale('log')
plt.savefig('issue2.png')
```

**Actual outcome**
Linear:
Logarithmic:

**Expected outcome**

Nice tick placement in both variants.

**Matplotlib version**
* Operating system: Arch Linux
* Matplotlib version (`import matplotlib; print(matplotlib.__version__)`): 3.4.1
* Matplotlib backend (`print(matplotlib.get_backend())`): Qt5Agg
* Python version: 3.9

Start by running the supplied reproduction with secondary_xaxis, first in the linear case and then with the logarithmic scale, to observe the overlapping tick positions. Trace how secondary_xaxis generates and transforms ticks; done means both variants produce readable, non-overlapping tick placement.

Written by the indexing model from the issue text.