fix(model): correct default-metric check in HIST and IGMTF

[genisis0x]

Description

Closes #2163.

metric_fn in qlib/contrib/model/pytorch_hist.py and qlib/contrib/model/pytorch_igmtf.py compares self.metric against a tuple with ==:

if self.metric == ("", "loss"):
    return -self.loss_fn(pred[mask], label[mask])

self.metric is a string (default ""), so self.metric == ("", "loss") is never true. The default / "loss" metric therefore falls through to raise ValueError("unknown metric"), breaking validation/early-stopping for these two models with the default config.

Every other torch model uses a membership test — e.g. pytorch_lstm.py:147, pytorch_alstm.py:151, pytorch_gru.py:151:

if self.metric in ("", "loss"):

Change

Change == to in in both metric_fn methods so the default metric returns the negative loss, matching the rest of the model zoo.

Testing

With metric="" (default) or "loss", metric_fn now returns -loss_fn(...) instead of raising. metric="ic" and unknown metrics are unaffected.

[genisis0x]

@SunsetWolf when you have a moment, would appreciate a review here. Small one-line-per-file fix: HIST and IGMTF used self.metric == ("", "loss") (a string-vs-tuple compare that's never true, so the default metric hit raise ValueError), switched to in to match pytorch_lstm/alstm/gru. CLA signed, CI green.