fix(bigquery): support @@ system variables in lexer

[bermi]

BigQuery supports both @param and @@system_variable syntax. The lexer regex only matched single @ prefixes, causing @@var to be split into two tokens. This produced unwanted line breaks during formatting (e.g. SET foo = \n@@bar;) and parse errors in CALL statements using system variables like @@error.message.

Widen the at_sign_literal regex from @[a-zA-Z_][\w]* to @{1,2}[a-zA-Z_][\w.]* to match both forms in a single token, with dot support for qualified names like @@error.message.

[benfdking]

Hey 👋

Thank you for the contribution! While this is correct, it would be better to distinguish between at_sign_literal and double_at_sign_literal, which have very different meanings.

In BigQuery, @param is a query parameter while @@error.message is a system variable — these are semantically distinct and SQLFluff treats them as separate token types.

SQLFluff introduced this in sqlfluff/sqlfluff#5307, where they define two separate lexer matchers:

RegexLexer(
    "at_sign_literal",
    r"@[a-zA-Z_][\w]*",
    ...
),
RegexLexer(
    "double_at_sign_literal",
    r"@@[a-zA-Z_][\w\.]*",
    ...
),

Along with a dedicated SystemVariableSegment that wraps the DoubleAtSignLiteralSegment, giving @@ variables their own node type in the AST.

It would be great to follow this pattern here — adding a separate DoubleAtSignLiteral syntax kind rather than combining both into a single @{1,2} regex. This keeps the token types semantically meaningful for downstream rules and analysis.

[bermi]

@benfdking thanks for the feedback. Let me know if I got the SystemVariableSegment right or if we need to handle it differently.