sky130: 46 failures with DRC now running (arrays, decoders, routing) #340

With the generic bitcell array path fixed (#336) and magic DRC actually running (#327), freepdk45 went from 48 failures to 11 and scn4m_subm from 45 to 11. sky130 reads 46. That is not new breakage: its DRC errors were previously reported as 0, and its LVS failures fell from 14 to 3.

Measured on dev at 86e8399, three technologies: 719 scheduled, 651 pass, 68 fail. ROM is excluded per #339.

sky130's 46

33  DRC
10  AssertionError
 3  LVS mismatch

Largest DRC counts

This is not the same cause as #336. Rules sampled on 15_local_bitcell_array_dummies_1rw_1r:

Metal3 minimum area < 0.24um^2                     (met3.6)
Metal2 spacing < 0.14um                            (met2.2)
All nwells must contain metal-connected N+ taps    (nwell.4)

#336 was nwell/pwell and nimplant/pimplant overlap caused by mirroring parity at abutted cell edges. These are spacing, minimum-area and well-tap rules, so the generic-path fixes should not be expected to carry over. The generic 15_local_bitcell_array_* equivalents now pass on freepdk45, which suggests the sky130 overrides in technology/sky130/custom/ have not had the same treatment.

Reproduce

cd compiler/tests && make -k -j<N> sky130
ls results/sky130/*.ok | wc -l ; ls results/sky130/*.bad | wc -l

Count the .ok and .bad stamps rather than trusting the exit status: make <target>.ok exits 0 even when a test fails.

For per-rule detail, magic's drc count total (what the harness parses) undercounts relative to drc listall why; load the kept .mag files and use the latter.