Add partition command to do balanced hypergraph partitioning on a gate-level module

[boqwxp]

This PR adds the partition command to do balanced hypergraph partitioning on a flattened gate-level module. Finding an optimal partitioning is generally an NP-hard problem, but fortunately there are good heuristics for it.

Miscellaneous changes:

This PR makes a few various enabling changes around the code base:

• Add a Design::select() interface for selecting whole modules (See PR rtlil: Add Design::select() for selecting whole modules #2178 )
• Add operator+=() to dict<>::iterator and dict<>::const_iterator and add operator+() and operator+=() to ObjIterator. This enables indexing into dicts with an iterator, like auto value = *(dict.begin() + offset), which helps reduce memory by avoiding the need for a node number -> cell dict<int, Cell *> (See PR hashlib, rtlil: Add operator+() and operator+=() to dict iterators #2177 )
• Break out statdata_t to the Yosys namespace. This is to support the -s option, where we need to measure the size of our module
• Add unit_gate_cost(), where all gates except NOT and BUF have a unit cost. This is the default cell type cost used for node weights in the hypergraph and for determining module size in the -s option

Solvers, interface, integration:

hMETIS seems to be the most performant solver, but has a very restrictive license. KaHIP has an MIT license but seems to have a nasty dependency on an old version of OpenMPI. minipart is also MIT licensed, but brings in boost dependencies.

@daveshah1 has indicated that he is developing hypergraph partitioning in-tree in nextpnr, so when that is done, I think we all agree to bring that in-tree, and I will update this PR and mark it as ready for review.

Until then, this command makes a temporary directory, writes a .hgr file (spec in this document), invokes minipart (which must be on the PATH), and reads a .sol file with the partition assignments.

Depends:

• PR log and qbfsat: Also include child process usage in PerformanceTimer::query() and report the time for each call to the QBF-SAT solver #2135
• PR hashlib, rtlil: Add operator+() and operator+=() to dict iterators #2177
• PR rtlil: Add Design::select() for selecting whole modules #2178

[whitequark]

The core data structure changes (iterator and selection) are good and can be merged right away as a separate PR.