[Spark] Fix type widening with char/varchar columns

[johanl-db]

Description

Using type widening on a table that contains a char/varchar column causes the following reads to fail with DELTA_UNSUPPORTED_TYPE_CHANGE_IN_SCHEMA:

CREATE TABLE t (a VARCHAR(10), b INT);
ALTER TABLE t SET TBLPROPERTIES ('delta.enableTypeWidening' = 'true');
ALTER TABLE t ALTER COLUMN b TYPE LONG;

SELECT * FROM t;
[DELTA_UNSUPPORTED_TYPE_CHANGE_IN_SCHEMA] Unable to operate on this table because an unsupported type change was applied. Field cut was changed from VARCHAR(10) to STRING`

Type changes are recorded in the table metadata and a check on read ensures that all type changes are supported by the current implementation as attempting to read data after an unsupported type change could lead to incorrect results.
CHAR/VARCHAR columns are sometimes stripped down to STRING internally, for that reason, ALTER TABLE incorrectly identify that column a type changed to STRING and records it in the type widening metadata.

The read check in turn doesn't recognize that type change as one of the supported widening type changes (which doesn't include changes to string columns).

Fix:

1. Never record char/varchar/string type changes in the type widening metadata
2. Never record unsupported type changes in the type widening metadata and log an assertion instead.
3. Don't fail on char/varchar/string type changes in the type widening metadata if such type change slips through 1. This will prevent failing in case a non-compliant implementation still record a char/varchar/string type change.
4. Provide a table property to bypass the check if a similar issue happens again in the future.

[cstavr]

LGTM! Thanks!