Currently, when HashME is applied, PSNR may drop (by several dB) for some TGM sequence pictures

[xujizheng]

When HashME is applied, early termination is used to skip some inter mode checking and can reduce encoding time.
However, the current design may skip inter mode too aggressively and cause PSNR decrease by several dB.
For example, frame 218@Desktop under LDB test condition, QP 22, there is ~8dB PSNR's difference comparing with neighboring frames (thanks Xin Zhao to provide the data).
POC 216 TId: 0 ( B-SLICE, QP 23 ) 10224 bits [Y 53.2713 dB U 45.5925 dB V 46.0552 dB]
POC 217 TId: 0 ( B-SLICE, QP 27 ) 26520 bits [Y 53.2250 dB U 45.0540 dB V 45.3901 dB]
POC 218 TId: 0 ( B-SLICE, QP 26 ) 27664 bits [Y 44.8762 dB U 45.1872 dB V 45.5623 dB]
POC 219 TId: 0 ( B-SLICE, QP 27 ) 10504 bits [Y 53.2358 dB U 45.5821 dB V 46.0491 dB]
POC 220 TId: 0 ( B-SLICE, QP 23 ) 15888 bits [Y 50.9829 dB U 45.5914 dB V 46.0576 dB]
POC 221 TId: 0 ( B-SLICE, QP 27 ) 27416 bits [Y 53.1610 dB U 45.1716 dB V 45.5504 dB]
It might influence CTC.
A fix is proposed as ​https://vcgit.hhi.fraunhofer.de/jvet/VVCSoftware_VTM/merge_requests/564 to address the issue.

[XiangLi]

Fixed as suggested.