Subset YAML parser errors do not name the config file

Context

Found while addressing the #64 review. load_config_file wraps parse errors from tomllib, json.loads and PyYAML as cannot parse config <path>: ..., but the built-in YAML subset parser (_parse_yaml_subset in src/autoforge/config.py) raises ConfigurationError directly, and load_config_file re-raises ConfigurationError unchanged. Those messages therefore never carry the file path:

YAML subset parser: bad indentation at: 'allow_merge: true' — install PyYAML for full YAML support
config must contain a mapping at top level          # subset parser, list at the root
config cfg.yaml must contain a mapping at top level # PyYAML, same file

An operator who passes --config from a wrapper script, or who keeps several config files, cannot tell from the message which file failed. The two backends also word the root-type error differently for the same input.

Proposed change

Either have _parse_yaml_subset raise a plain ValueError so the existing cannot parse config <path>: wrapper applies (the duplicate-key error added in #64 already does this), or prefix the path in load_config_file when re-raising a ConfigurationError from a parser. Tests that match= on the message body keep passing either way since pytest.raises(match=) uses re.search.

Cosmetic; no behavioural change to what is accepted or rejected.

Start in src/autoforge/config.py by reading _parse_yaml_subset and load_config_file, then inspect the existing parser-error tests and their match= assertions. Make subset-parser failures identify the config path consistently with the other backends, without changing which inputs are accepted or rejected, and run the relevant configuration tests.