Issue Description

Arrow does not seem to perform validation on timestamps for unusual formats where the information conflicts.

The following code snippet runs just fine:

>>> arrow.get("2021-01-30 14:00:00 AM", "YYYY-MM-DD hh:mm:ss A")
<Arrow [2021-01-30T14:00:00+00:00]>

First of all, since hh is documented to go to a maximum value of 12, I expect arrow to raise an err

Is your feature request related to a problem? Please describe.
I would be useful to have the ability to make a reverse copy of a column and/or table.
I am thinking of std::reverse (https://en.cppreference.com/w/cpp/algorithm/reverse) . But not necessarily using iterators, since that might make things more complicated.

Describe the solution you'd like
The following APIs could be

We no longer need to control the number of concurrent kernels, since now we control the number of concurrent tasks

TurboDBC is probably the fastest method to communicate to a db with arrow data. We can implement utility functions for this in Python.

If turbodbc isn't installed we raise an exception. It probably has to be installed with conda, so we cannot add it as optional dependency.

All required arrow interop functions are already available to make this work.

It's been reported by other projects that musl could lead to performance issues, so we should switch our Linux binary releases to use glibc instead to avoid these kind of potential pitfalls.

Problem description

Reading a dataset with eager's read functionality raises a ValueError when providing columns.

Example code (ideally copy-pastable)

import pandas as pd

from tempfile import TemporaryDirectory
from functools import partial
from storefact import get_store_from_url

from kartothek.io.eager import store_dataframes_as_dataset, read_dataset_as_data

It would be helpful to have Fletchgen output warnings for unused metadata fields that start with fletcher_. For example, (this happened to me) when someone adds fletchgen_epc to Schema metadata instead of Field metadata.