[Spark] Remove type widening metadata

[johanl-db]

Description

Type widening records type changes applied to a table in the table schema using the following metadata:

"metadata": {
  "delta.typeChanges": [{
    "toType": "short",
    "fromType": "byte"
  }]

The initial intent was to keep that metadata in schemas returned to users, e.g. via df.schema, as it may provide useful information.
That is a bad idea though:

• This allows type widening metadata to leak outside of the table, possibly to other tables. While this isn't expected to cause correctness issues - the type widening metadata is mostly informative - it would still be confusing.
• It is ultimately internal Delta metadata that shouldn't be surfaced to users. Other features, such a column mapping, explicitly remove their related metadata before surfacing schemas to users.

How was this patch tested?

• Updated existing tests to remove type widening metadata when checking for returned schema.
• Added test to ensure we don't leak type widening metadata

Does this PR introduce any user-facing changes?

Type widening metadata that was visible in dataframe schemas isn't surfaced anymore:
df.schema:
Before:

"fields": [{
  "name": "a",
  "type": "integer",
  "nullable": true,
  "metadata": {
    "delta.typeChanges": [{
      "toType": "integer",
      "fromType": "short"
    }]
  }
}]

After:

"fields": [{
  "name": "a",
  "type": "integer",
  "nullable": true,
  "metadata": {}
}]