fix(spark): parse SQL scripting compound blocks

[judahrand]

This is a proposed fix to #7838. I appreciate that this does not fundamentally fix the underlying lack of mature support for imperative SQL. However, it does allow SQL script blocks to roundtrip. I can also understand that this doesn't really solve the issue for transpiling between dialects which is a core sqlglot functionality. So based on that please feel free to close - I figured the approach might spark some idea in someone else.

[georgesittas]

I don't think this is the right approach:

• What does "wrapped" represent and why does it need to be an AST argument? It'd seem more reasonable to me to abstract the BEGIN, END delimiters away (i.e., remove from the AST) and only emit them at generation time if a dialect requires them to be present.
• Overriding the parsing for TokenType.BEGIN in Spark may have side-effects in Databricks, which supports BEGIN {TRANSACTION | WORK}.

Based on Spark's grammar, I think we should override the statement parser for Spark and map it to _parse_block, making sure Databricks pops it, or that it overrides it further to ensure its own statements starting with BEGIN can still be parsed.