Handle near zero coefficients in tcq

[jingninghan]

Add functions to handle the coefficients that are quantized to 1.
In such cases, there is no need to check upper abs levels in the
trellis framework, and hence reducing the complexity.

Anchor: b5bb354

Test condition: CTC, RA, 33 frames, speed 1

+---------+--------+--------+--------+--------+----------+
| Summary |   Y    |   U    |   V    |  YUV   | Enc-time |
+---------+--------+--------+--------+--------+----------+
| A1      |  0.03% |  0.06% |  -0.12% |  0.02% | 96.12%   |
| A2      |  0.06% |  0.31% |  -0.19% |  0.06% | 96.18%   |
| A3      |  0.11% |  -0.21% | -0.42% |  0.06% | 97.11%   |
| A4      |  0.16% |  0.76% |  -0.91% |  0.14% | 97.20%   |
| A5      | -0.12% | -0.17% | -1.18% | -0.20% | 96.32%   |
| B1      | -0.12% |  0.03% |  0.03% | -0.11% | 96.15%   |
|avg wo b2|  0.03% |  0.16% | -0.32% |  0.01% | 96.42%   |
+---------+--------+--------+--------+--------+----------+
STATS_CHANGED

[joeyoung-yt]

Is there a description for this change (or bug#)? How about coding gain?

[jingninghan]

Is there a description for this change (or bug#)? How about coding gain?

Added both. ptal thanks

[joeyoung-yt]

I'm curious about the BDrate change. It's minor, but I'd expect it to be closer to neutral.
Is it due to the q1 change (checking just 0 and 1)?
The speedup looks good, with just a minor change in bdrate.

[jingninghan]

e q1 change (ch

Yes, i think that's mainly due to constraining it to only 0 and 1 levels.