Extend _json_compat with load/dump and migrate CUTracer to orjson

Summary:
Extend tritonparse's `_json_compat.py` compatibility layer with file-based
`load(f)` and `dump(obj, f)` convenience wrappers, then migrate all 14 CUTracer
production Python files from stdlib `json` to use `tritonparse._json_compat`.

This gives CUTracer a free performance upgrade via orjson for JSON parsing and
serialization (3-10x faster), while maintaining graceful degradation to stdlib
json in environments where orjson is unavailable.

Key changes:
- `_json_compat.py`: Add `load()` and `dump()` functions that delegate to the
  existing `loads()`/`dumps()` with file I/O wrapping
- CUTracer: Replace `import json` with `from tritonparse._json_compat import ...`
  across query/, analyze/, reduce/, and validation/ modules
- No changes to test files (consistent with tritonparse convention)
- No changes to `scripts/parse_instr_hist_trace.py` (standalone, no tritonparse dep)

CUTracer already depends on tritonparse (Buck + pip), so no new dependencies needed.

___

overriding_review_checks_triggers_an_audit_and_retroactive_review
Oncall Short Name: triton

Differential Revision: D99939361

fbshipit-source-id: 1ad5f9bce150e0aba8d7f55406e217d52cb923d2

Author: FindHao

tritonparse/_json_compat.py

@@ -9,6 +9,7 @@
 
 ``loads()`` accepts ``str | bytes | bytearray | memoryview`` inputs.
 ``dumps()`` returns ``str``.
+``load()`` and ``dump()`` provide file-like object I/O convenience wrappers.
 """
 
 try:
@@ -85,3 +86,20 @@ def dumps(obj, *, indent=False, sort_keys=False):
         if sort_keys:
             kwargs["sort_keys"] = True
         return _json.dumps(obj, **kwargs)
+
+
+def load(f):
+    """Deserialize a file-like object containing JSON to a Python object."""
+    return loads(f.read())
+
+
+def dump(obj, f, *, indent=False, sort_keys=False):
+    """Serialize a Python object as JSON to a file-like object.
+
+    Args:
+        obj: The object to serialize.
+        f: A file-like object with a ``write()`` method.
+        indent: If True, pretty-print with 2-space indent.
+        sort_keys: If True, sort dictionary keys.
+    """
+    f.write(dumps(obj, indent=indent, sort_keys=sort_keys))