Net decomposing router: hotfix

[duck2]

Looks like sinks no longer "grow" to the size of their tile. In that case, it's better to use the tile bounding box to sample sinks for the net decomposing router.

This seems to fix the runtime blowup I started to see again with the decomp + flat router. Here is a table with "actual" route times from the flat router which I get by adding up the iteration times.

circuit	master	mine	decomp	decomp (fixed)
gsm_switch_stratixiv_arch_timing.blif	113.7	108	72.2	63.6
mes_noc_stratixiv_arch_timing.blif	467.7	254.6	733.6	655.4
dart_stratixiv_arch_timing.blif	68.5	68.2	47.8	42.7
denoise_stratixiv_arch_timing.blif	219.4	277.8	97.2	92.4
sparcT2_core_stratixiv_arch_timing.blif	173.8	187.4	114.4	109.5
cholesky_bdti_stratixiv_arch_timing.blif	115.9	121.8	83.3	65.8
minres_stratixiv_arch_timing.blif	69.7	69.7	38.2	35.4
stap_qrd_stratixiv_arch_timing.blif	71.5	65	147.2	47.6
openCV_stratixiv_arch_timing.blif	112.1	106.1	78.3	65.6
bitonic_mesh_stratixiv_arch_timing.blif	137.3	131.3	73.9	72
segmentation_stratixiv_arch_timing.blif	117.5	112.1	70.7	66.1
SLAM_spheric_stratixiv_arch_timing.blif	80.1	78.2	59.7	57
des90_stratixiv_arch_timing.blif	72.4	74.8	1015.1	102.9
neuron_stratixiv_arch_timing.blif	28.2	25.8	17.8	16.1
sparcT1_core_stratixiv_arch_timing.blif	48.7	45.8	34.6	32.5
stereo_vision_stratixiv_arch_timing.blif	20.9	23.6	17.2	16.9
cholesky_mc_stratixiv_arch_timing.blif	39.1	38.1	27.1	25
directrf_stratixiv_arch_timing.blif	710.4	520.7	385.6	364.2
bitcoin_miner_stratixiv_arch_timing.blif	592.1	577	477.1	448.3
LU230_stratixiv_arch_timing.blif	416.5	418.4	453.6	287.2
sparcT1_chip2_stratixiv_arch_timing.blif	255	270.2	178.8	169.5
LU_Network_stratixiv_arch_timing.blif	180.4	189.9	115.4	116.4

[vaughnb-cerebras]

The code looks fine and the results are improved so merging. But you should add a comment explaining these two functions are used for sinks, and hence you want to check how close the block they are part of is to the cutline (hence the tile code).