Clarify query-backed Dataset.source representation and SQL dialect

[mchades]

Problem

The core specification
defines Dataset.source as a string that may identify a physical table/view or contain a query.

For a query-backed source, the specification provides neither:

• a discriminator that distinguishes a relation identifier from SQL text; nor
• a SQL dialect for parsing or validating the query.

Consumers therefore cannot interpret query-backed sources consistently.

Questions

1. Is a query-backed Dataset.source intended to be portable core behavior?

2. If so, should the specification provide a structured form such as:

   source:
     type: query
     dialect: ANSI_SQL
     query: SELECT ...

3. Should the existing string form remain the relation-identifier form for compatibility?

4. If query interpretation is intentionally consumer-specific, should the specification
   explicitly allow consumers to reject query-backed sources?

Context

Gravitino maps an OSI dataset source to a catalog Table or View. Because the query dialect is
currently unspecified, its proposed structured Metric View model requires a query to be registered
as a logical View before it can be referenced as a dataset source.

Related discussions: #52, #294, and #104.