Avoid moving back the lexer for triple-quoted fstring (#11939)

## Summary

This PR avoids moving back the lexer for a triple-quoted f-string during
the re-lexing phase.

The reason this is a problem is that for a triple-quoted f-string the
newlines are part of the f-string itself, specifically they'll be part
of the `FStringMiddle` token. So, if we moved the lexer back, there
would be a `Newline` token whose range would be in between an
`FStringMiddle` token. This creates a panic in downstream usage.

fixes: #11937 

## Test Plan

Add test cases and validate the snapshots.

crates/ruff_python_parser/resources/invalid/re_lexing/triple_quoted_fstring_1.py

@@ -0,0 +1,6 @@
+# There are trailing whitespace before the newline character but those whitespaces are
+# part of the comment token.
+# https://github.com/astral-sh/ruff/issues/11929
+
+f"""hello {x # comment    
+y = 1

crates/ruff_python_parser/resources/invalid/re_lexing/triple_quoted_fstring_2.py

@@ -0,0 +1,6 @@
+# The lexer can't be moved back for a triple-quoted f-string because the newlines are
+# part of the f-string itself.
+# https://github.com/astral-sh/ruff/issues/11937
+
+f'''{foo:.3f
+'''

crates/ruff_python_parser/resources/invalid/re_lexing/triple_quoted_fstring_3.py

@@ -0,0 +1,7 @@
+# Here, the nesting level is 2 when the parser is trying to recover from an unclosed `{`
+# This test demonstrates that we need to reduce the nesting level when recovering from
+# within an f-string but the lexer shouldn't go back.
+
+if call(f'''{x:.3f
+'''
+    pass