API: better error-handling for df.set_index

[h-vetinari]

splitting up #22236.

Let's have
df = pd.DataFrame(np.random.randn(5, 5), columns=list('ABCDE'))

The error handling of df.set_index can be improved in at least three cases:

1. df.set_index(['A', 'A'], drop=False) works, while
   df.set_index(['A', 'A'], drop=True) yields
   KeyError: 'A'
2. Objects of unknown type yield KeyError instead of TypeError:
   df.set_index(map(str, df.A))
KeyError: "None of [Index([...], dtype='object')] are in the [columns]"
3. df.set_index(['foo', 'bar', 'baz']) only shows one missing key
   KeyError: 'foo' (in a huge stacktrace)

Better would be:

1. gracefully handle duplicate column names when drop=True
2. raise better error message, e.g. TypeError: only allowed types are: ...
3. Show all missing keys: KeyError: "['foo', 'bar', 'baz']"