Hermitian pinv in simultaneous inpainting

I've noticed some instances of pinvs not working because an SVD doesn't converge, but adding this option seems to avoid that (not sure why). It's also supposed to be slightly faster!

hera_cal/lst_stack/averaging.py

@@ -511,7 +511,7 @@ def average_and_inpaint_simultaneously_single_bl(
                     / base_noise_var[night_index, band],
                     basis,
                 )
-                _CNinv_1sample_dpss_inv = np.linalg.pinv(_CNinv_1sample_dpss)
+                _CNinv_1sample_dpss_inv = np.linalg.pinv(_CNinv_1sample_dpss, hermitian=True)
                 CNinv_1sample_dpss.append(_CNinv_1sample_dpss)
                 CNinv_1sample_dpss_inv.append(_CNinv_1sample_dpss_inv)
                 cache[hash_key] = (_CNinv_1sample_dpss, _CNinv_1sample_dpss_inv)
@@ -542,7 +542,7 @@ def average_and_inpaint_simultaneously_single_bl(
         is_unflagged_night = ~np.all(stackf[:, band], axis=1)
 
         # Compute weighted sample mean from per-day DPSS-fits and noise-weighted covariance matrix
-        inv_sum_CNinv_dpss = np.linalg.pinv(sum_CNinv_dpss)
+        inv_sum_CNinv_dpss = np.linalg.pinv(sum_CNinv_dpss, hermitian=True)
         sample_mean_dpss = inv_sum_CNinv_dpss @ np.einsum(
             "nde,nd->e", CNinv_dpss[is_unflagged_night], dpss_fits[is_unflagged_night]
         )