More fixes to sync_netlists_to_routing_flat

[fkosar-ql]

This PR contains a number of fixes to enable OpenFPGA bitstream generation using flat router results:

• "Unset" port equivalence in the architecture (in memory) before flat routing starts. This is for compatibility with OpenFPGA (also may simplify the sync code)
• Rename new cluster nets created while rebuilding the clustered netlist more consistently (avoid name clashes)
• In annotate_rr_node_nets, don't fail right away if two cluster nets point to the same RR nodes, instead check if they are mapped to the same atom net. (This lets OpenFPGA have net -> rr node annotations without having to split the flat route trees if an atom net produced two or more cluster nets)

[vaughnbetz]

Crash in flat routing so some bug still.
Please add an output message (that is easy to interpret) when you turn off pin equivalence in memory (should say you're doing this and why).
Should run the nightly tests manually once basic CI passes.

[vaughnbetz]

Looks mostly good -- some commenting requested.

[fkosar-ql]

No idea how it's passing with the current version of sync_netlists_to_routing_flat, but spree.v seems to connect some global nets to CLBs and routing fails when I try to run with --constant_net_method route. This means the synchronized results will be invalid since the flat router does not see which pins the globals are using.

For now I'm just adding a warning for that case to make the CI work, but it should be handled at some point.