Failure to convert nested object raises exception instead of warning

[emolter]

Attempting to open an asdf file that requires a converter to deserialize some portion of the file raises an exception when that converter is not found, e.g. if the parent package decided to drop support for it. It might be more helpful if this raised a warning and loaded the un-convert-able parts as TaggedDict.

[braingram]

One option here might be to not allow converter errors to halt opening a file. Let's say there's a file:

#ASDF 1.0.0
#ASDF_STANDARD 1.6.0
%YAML 1.1
%TAG ! tag:stsci.edu:asdf/
--- !core/asdf-1.1.0
quantity: !unit/quantity-1.2.0
  unit: !unit/unit-1.0.0 m
  value: !core/ndarray-1.1.0
    data: [1, 2]
    datatype: float64
    byteorder: little
    shape: [2, 3]
...

Here shape doesn't match data, which triggers an exception. This is just an example of a converter that creates an exception.
On asdf main (with asdf-astropy) opening this causes the exception and halts. There is no opportunity here to inspect the data (or if there was other correctly serialized data that would also not be accessible).
However, if we catch the exception in the proxy and convert it into a warning (type and message TBD) we can change the ndarray exception (and resulting quantity exception) into warnings allowing the data to still be accessible.

/Users/bgraham/projects/src/asdf/asdf/extension/_converter.py:305: UserWarning: Converter <asdf._core._converters.ndarray.NDArrayConverter object at 0x102fae810> failed to deserialize: inline data doesn't match the given shape
  warnings.warn(f"Converter {self._delegate} failed to deserialize: {err}", UserWarning)
/Users/bgraham/projects/src/asdf/asdf/extension/_converter.py:305: UserWarning: Converter <asdf_astropy.converters.unit.quantity.QuantityConverter object at 0x1029e8320> failed to deserialize: The value must be a valid Python or Numpy numeric type.
  warnings.warn(f"Converter {self._delegate} failed to deserialize: {err}", UserWarning)
{'quantity': {'unit': Unit("m"), 'value': {'data': [1, 2], 'datatype': 'float64', 'byteorder': 'little', 'shape': [2, 3]}}}

[braingram]

To expand on this a bit. I propose that we consider.

1. Add a warn_on_failed_conversion option to AsdfConfig
2. Set the default option of above to False (to retain backwards compatibility)
3. when above is enabled, catch converter errors and turn them into warnings

This would allow a user to:

import asdf
asdf.config.get_config().warn_on_failed_conversion = False
tree_with_broken_data = asdf.load("broken_file")
# inspection of the tree is possible

[emolter]

As we discussed in person, that plan sounds great to me. Let me know if you need any support from me with that.