Escape the APPLY column transformer's function name and column-name prefix when formatting - #119851

…refix when formatting

`ASTColumnsApplyTransformer::formatImpl` wrote both of its string members as raw bytes:

    ostr << func_name;
    ostr << ", '" << column_name_prefix << "')";

The parser reads them back as an identifier (`ParserIdentifier`) and as a string literal
(`ParserStringLiteral`), so any byte that is special to those grammars ends the token early or
is re-interpreted on the way in. Formatting is one half of a round-trip contract:
`executeQueryImpl` formats every query and parses it back under `#ifndef NDEBUG`, gated by no
setting, and raises `LOGICAL_ERROR` when the result does not match.

Three plain SELECTs therefore aborted an assertions-enabled build, all measured on a Debug
build of 305623c:

    SELECT * APPLY (toString, 'x''y') FROM (SELECT 1 AS a)
      -> "cannot parse query back", signal 6
    SELECT * APPLY (toString, 'a\\b') FROM (SELECT 1 AS a)
      -> the prefix is one backslash, spelled '\\' in SQL; emitted raw it re-parses as a<BS>b,
         so "the original AST ... differs from the result of parsing back formatted AST",
         signal 6
    SELECT * APPLY `to String` FROM (SELECT 1 AS a)
      -> "cannot parse query back", signal 6

The same formatter runs with no assertions, so release builds are affected through
`formatQuery` / `formatQuerySingleLine`, which returned SQL that either does not parse or
parses to a different query. The backslash case was not even idempotent: re-formatting mutated
the prefix again. A prefix holding a newline additionally broke `formatQuerySingleLine`'s
single-line contract by emitting a literal newline.

`quoteString` and `backQuoteIfNeed` are the exact inverses of those two parsers, and are what
every sibling in this area already uses: `ASTColumnsExceptTransformer::formatImpl` quotes its
pattern, `Replacement::formatImpl` back-quotes its name, and `ASTColumnsMatcher` quotes its
patterns. `backQuoteIfNeed` leaves an ordinary function name untouched and `quoteString("p_")`
is `'p_'`, so the output is byte-identical for every value that is not already broken and no
existing reference changes.

`appendColumnName` writes the same two members raw a few lines below and is deliberately left
alone: its output is a column name that is never parsed back, so quoting it would rename
user-visible columns for exactly the prefixes containing a quote or a backslash, which work
correctly in release builds today.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>