Verilog: assignments between unpacked arrays of equivalent types

[kroening]

Fixes the KNOWNBUG from #2099. Based on #2099, please merge that one first.

Per 1800-2017 section 7.6, an unpacked array can be assigned an unpacked array of
an equivalent type, and section 6.22.1 gives the equivalence rules:

• (c) fixed-size unpacked arrays are equivalent when they have the same size and
  equivalent element types;
• (d) packed arrays, packed structs, packed unions and the built-in integral
  types are equivalent when they have the same number of bits, are all 2-state or
  all 4-state, and are all signed or all unsigned.

An unpacked array of an eight-bit packed struct is therefore assignable to, and
connectable to a port of, an unpacked array of [7:0].

Changes

• src/verilog/verilog_typecheck_expr.cpp — new equivalent_type
  implementing 6.22.1, with is_integral_type (6.11.1) and is_signed_type
  (7.2.1, 7.4.1) helpers. assignment_conversion now inserts a typecast where
  the two unpacked array types are equivalent but not identical, instead of
  rejecting the assignment.
• src/verilog/verilog_lowering.cpp — the lowering of a typecast to a
  struct, union or array passed the operand to from_bitvector without
  converting it to a bit vector first, which tripped the src.type().id() != ID_array precondition in extract once the operand was itself of a composite
  type. It now goes through to_bitvector, which returns non-composite operands
  unchanged, so the existing paths are unaffected.

Note that the elaborated types do not record whether a vector was declared as
2-state or 4-state (logic [7:0] and bit [7:0] both elaborate to the same
type), so the 2-state/4-state condition only rules out the types that carry x
and z. This is noted in a comment.

Tests

• structs/array_of_structs2 — flipped KNOWNBUG to CORE (the reported port
  connection).
• structs/array_of_structs3 — new CORE test asserting the conversion in both
  directions, checking the order of the array elements and the order of the
  struct members, using a two-field struct against [15:0] so a swap would be
  caught.
• structs/array_of_structs4 — new CORE test that non-equivalent element widths
  are still rejected.

Also checked by hand, and still rejected: array size mismatch, signedness
mismatch, and an unpacked struct as the element type. Nested
array of array of struct works via the recursive case.

Full regression/{verilog,ebmc,smv,vlindex} in -C and -K modes plus the unit
tests pass.