Prune BAWP by GOP usage at speed >= 4

[xinxiong-xx]

Prune BAWP by GOP usage at speed >= 4

At speed 4 the encoder searches BAWP on almost every inter frame. Blocks at high pyramid levels rarely choose it, yet its search takes a similar share of the encode time at every level.

This change measures, on each GOP's level-1 and level-2 frames, the share of the BAWP-eligible luma area whose final block uses BAWP. If the mean of the two shares is at most 0.5, BAWP is turned off on the GOP's frames at pyramid level >= 3. Intra and level-1 frames start a new GOP, frames at levels 1 and 2 are never pruned, and only frames coded earlier are used. The speed feature is prune_bawp_by_gop_usage.

STATS_CHANGED for speed >= 4.

CTC RA, A1 and A2, 33 frames (Anchor: 72fd5a2)

speed = 4

Class	Y	Cb	Cr	wAvg	Enc%	Dec%
A1	-0.00	-0.02	+0.15	+0.00	98.5	100.6
A2	+0.00	-0.00	-0.03	+0.00	98.2	100.5

speed = 5

Class	Y	Cb	Cr	wAvg	Enc%	Dec%
A1	+0.05	+0.17	+0.20	+0.06	98.8	99.9
A2	-0.01	-0.09	+0.03	-0.01	98.0	100.1

speed = 6

Class	Y	Cb	Cr	wAvg	Enc%	Dec%
A1	+0.00	+0.14	+0.12	+0.01	97.7	100.2
A2	-0.01	-0.19	-0.14	-0.02	96.8	99.4

[leolzhao]

Good trade off!