Allow py_wheel to use analysis phase information for attributes

# 🚀 feature request

### Relevant Rules

`py_wheel`

### Description

When building a wheel with `py_wheel`, some attributes are closely tied to the Python runtime the wheel is intended for. For example, building a wheel with a C extension can require a `python_tag` value that has the python version as part of the abi tag.

Because e.g. `py_wheel.python_tag` takes a string, the only way to pass a value is to use a select() expression to map some condition to a raw string. This is tedious, to say the least. It's also problematic for constructing the `cp39` string -- there's no flag that captures the runtime implementation name ("cp"), so creating a select for this is tough (though, still possible with some tricks, see below).

I _think_ a workaround here is to combine toolchain lookup with FeatureFlagInfo. Have a custom rule look up the toolchain and return e.g. implementation_name in FeatureFlagInfo. Have a config_setting map that to true/false. Have a select() expression map the config setting to e.g. "cp39". Complicated and verbose, but I think it would work.

### Describe the solution you'd like

It'd be better if py_wheel either inferred the values from the toolchain, or provided a way to feed analysis phase information in for value.

The loose idea I have is to have an label attribute that accepts e.g. PyWheelInfo provider, and then users can populate that with what they choose. We'd provide a default impl that did the common thing.

Beginne bei der py_wheel-Regel und konzentriere dich auf python_tag sowie die hier beschriebene Toolchain-Information aus der Analysephase. Ermittle, ob Inferenz oder ein PyWheelInfo-Eingabewert Laufzeit-abgeleitete Attribute besser unterstützt; erledigt ist die Aufgabe, wenn Benutzer Werte wie das Implementierungs-/Versions-Tag ohne den derzeitigen ausführlichen select()-Workaround angeben können.

Vom Indexierungsmodell aus dem Issue-Text verfasst.