Read hdf returns unexpected values for categorical

[nofarm3]

• [V] closes BUG: queries on categorical string columns in read_hdf return unexpected results #39189
• [V] tests added / passed
• [V] Ensure all linting tests pass, see here for how to run them
• [V] whatsnew entry

This bug happens when filtering on categorical string columns and choose a value which doesn't exist in the column.
Instead of returning an empty dataframe, we get some records.
It happens because of the usage in numpy searchsorted(v, side="left") that find indices where elements should be inserted to maintain order (and not 0 in case that the value doesn't exist), like was assumed in the code.
I changed it to first the for the value, and use searchsorted only if value exists, I also added a test for this specific use case.
I think in the long run, maybe we should refactor this area in the code since one function covers multiple use cases which makes it more complex to test.

In addition, I moved the logic to a new method to keep the single-responsibility principle and to make it easier to test.

[jreback]

note that you don't need to open a new PR, you can simply push to the original one. since you already did, then ok.

[nofarm3]

I think the failures here are not related to my changes.

[jreback]

thanks @nofarm3