fix: handle named parameters in identifier contexts

[edjubert]

What kind of change does this PR introduce?

Bug fix.

What is the current behavior?

Named parameters are normalized before parsing, but parameters used as SQL
identifiers need a synthetic identifier placeholder because PostgreSQL does not
accept positional parameters in relation or qualifier positions.

For example, :raw_data.documents is normalized to a parser-compatible form
such as a.documents. Parsing succeeds, but typechecking then reports the
false positive relation "a.documents" does not exist.

Fixes #787.

What is the new behavior?

Named parameters are classified by SQL context:

• value parameters continue to become positional parameters;
• identifier parameters, including qualified names such as
  :raw_data.documents, become length-preserving synthetic identifiers for
  parsing;
• PostgreSQL typechecking is skipped only for the statement containing an
  unresolved named identifier parameter.

Parsing, syntax diagnostics, linting, and plpgsql_check remain enabled.
Other statements in the same file continue to be typechecked normally.

Additional context

The conversion keeps the original byte length so parser diagnostic offsets stay
aligned with the source document.

The change does not attempt to resolve external aliases such as raw_data.
A future dbsh integration may provide that semantic context.