The validate command already accepted a --dialect flag, but parse,
format, analyze, and lint were hard-wired to the default (PostgreSQL)
dialect and rejected dialect-specific SQL such as MySQL's
`LIMIT offset, count`.

Add a --dialect flag to each of the four commands, validate the value
up front (unknown names are rejected), and thread it into the
tokenizer and parser at the point each command tokenizes/parses.

Shared helpers live in cmd_utils.go: validateDialectName, plus
tokenizerForDialect and parserForDialect. For an explicit dialect a
fresh, non-pooled tokenizer is created rather than reconfiguring a
pooled one, because tokenizer.Reset does not clear dialect/keyword
state and a pooled dialect-configured tokenizer would leak that state
to later callers.

The linter package gains an exported SetDialect method so lint can
tokenize/parse dialect-specific syntax, and lint --security now parses
with the selected dialect via gosqlx.ParseWithDialect.

The default remains PostgreSQL, so behavior without --dialect is
unchanged.

Fixes: #523

# Conflicts:
#	.trivyignore