Individual tensor metadata object query tool

```
$ python3 safetensors_m.py ss_network_module /path/to/file/summer_dress.safetensors
"networks.lora"

$ python3 safetensors_m.py nonexistent_module /path/to/file/summer_dress.safetensors
Error: Metadata does not contain a `nonexistent_module` item, did you spell it right?

$ python3 safetensors_m.py ss_network_module /path/to/file/weird_file.safetensors
Error: File header does not contain a `__metadata__` item

$ python3 safetensors_m.py ss_tag_frequency /path/to/file/trina.safetensors
{
    "6_trina": {
        "trina": 26, 
        " black hair": 20, 
        " hands on hips": 1, 
        " looking at viewer": 7
    }
}
```

This is a proof of concept utility for retrieving the value of a named object inside `__metadata__` from a safetensors file, with the intention of eventually moving the code into `safetensors_util.py` as its own separate `cli.command()` block. I wrote this because so far I haven't seen any other utilities for retrieving these values from safetensors files, e.g. checking `ss_network_module` to tell whether a file is LoRA (`networks.lora`) or Lycoris (`lycoris.kohya`) before trying to install it into a potentially incompatible SD client, or just debugging troublesome safetensors files without having to dump/parse the entire metadata block. Applies `_ParseMore()` to all output for readability's sake but this could be optionally disabled.

safetensors_object.py

@@ -0,0 +1,23 @@
+import json, sys
+from safetensors_file import SafeTensorsFile
+from safetensors_worker import _ParseMore
+
+def get_object(tensorsfile: str) -> str:
+	s = SafeTensorsFile.open_file(tensorsfile, quiet=True)
+	js = s.get_header()
+	s.close_file()
+
+	if "__metadata__" not in js:
+		return "Error: File header does not contain a `__metadata__` item"
+	md = js["__metadata__"]
+	if md_object not in md:
+		return f'Error: Metadata does not contain a `{md_object}` item, did you spell it right?'
+	_ParseMore(md) # pretty print the metadata
+	stf = md[md_object]
+	return json.dumps(stf, ensure_ascii=False, separators=(', ', ': '), indent=4)
+
+md_object = sys.argv[1]
+tensorsfile = sys.argv[2]
+hdata = get_object(tensorsfile)
+
+print(hdata)