Backslashes are wrongly escaped in grammar_codegen

When I change [fuzztest/grammars/JSON.g4](https://github.com/google/fuzztest/blob/main/fuzztest/grammars/JSON.g4) to support some escape sequences
```antlr
STRING : '"' (CHARACTER | '\\' ESCAPED )* '"' ;

ESCAPED : '"' | '\\' | '/' | 'b' | 'f' | 'n' | 'r' | 't' ;
```
then the grammar tests fail because the nlohmann/json library fails to parse the generated json:
```
$ bazel test //domain_tests:in_grammar_domain_test
...
Executing tests from //domain_tests:in_grammar_domain_test
-----------------------------------------------------------------------------
Running main() from gmock_main.cc
[==========] Running 6 tests from 2 test suites.
[----------] Global test environment set-up.
[----------] 5 tests from InJsonGrammar
[ RUN ] InJsonGrammar.InitGeneratesDifferentValidJson
domain_tests/in_grammar_domain_test.cc:41: Failure
Value of: parsed_json.is_discarded()
Actual: true
Expected: false
"{\"\\\\\"\":[80.0E3,false]}"
...
```
That's because backslashes are escaped when they shouldn't be. E.g. `"\""` (quote, escaped quote, quote) becomes `"\\""` (quote, escaped backslash, quote, quote), which is not correct json.

The linked PR fixes the issue.