Replace YAJL with c-json for JSON parsing and generation

[giuseppe]

Alternative to #170

[jnovy]

Concern 1: json_object_new_string_len truncation:

In json_gen_string(), len is size_t but cast to (int) for json_object_new_string_len. For strings > 2GB this would silently truncate. Practically impossible in OCI spec contexts, but a SIZE_MAX > INT_MAX guard or comment would be defensive.

Concern 2: Residual handling semantic change

The old code stored residuals as a yajl_val tree (preserving exact in-memory structure). The new code serializes residuals to a JSON string (char *) and re-parses when generating output. This adds a serialize/parse round-trip. In practice json-c handles this well, but it's worth noting as a behavioral change - e.g., key ordering within residuals may change.

Concern 3: clang-format auto-formatting may cause inconsistency

generate.py now runs clang-format -i on generated files if clang-format is available. This means the generated output differs depending on whether clang-format is installed, which could cause confusion in CI or when comparing outputs across environments. Consider making this opt-in via a flag.

I'm happy we lean towards json-c here!

[jnovy]

Memory leak:

// Current code:
for (i = 0; i <= g->depth; i++)
  {
    free (g->pending_key[i]);
  }

// Should be:
for (i = 0; i <= g->depth; i++)
  {
    free (g->pending_key[i]);
    json_object_put (g->stack[i]);
  }

[giuseppe]

Concern 1: json_object_new_string_len truncation:

In json_gen_string(), len is size_t but cast to (int) for json_object_new_string_len. For strings > 2GB this would silently truncate. Practically impossible in OCI spec contexts, but a SIZE_MAX > INT_MAX guard or comment would be defensive.

Concern 2: Residual handling semantic change

The old code stored residuals as a yajl_val tree (preserving exact in-memory structure). The new code serializes residuals to a JSON string (char *) and re-parses when generating output. This adds a serialize/parse round-trip. In practice json-c handles this well, but it's worth noting as a behavioral change - e.g., key ordering within residuals may change.

Concern 3: clang-format auto-formatting may cause inconsistency

generate.py now runs clang-format -i on generated files if clang-format is available. This means the generated output differs depending on whether clang-format is installed, which could cause confusion in CI or when comparing outputs across environments. Consider making this opt-in via a flag.

I'm happy we lean towards json-c here!

thanks!

Addressed now

[jnovy]

LGTM