igla race: backend.t27 and rtl.t27 import each other -- the only import cycle in specs/ #10682

The corpus run of the dependencies view (#10680, PR #10681) found the only import cycle in specs/. On master e607228, specs/igla/race/backend.t27 line 12 has use igla::race::rtl;, and specs/igla/race/rtl.t27 line 11 has use igla::race::backend;. tsort over t27c use-edges --specs-dir specs reports the same loop on stderr.

A mutual import makes the order of the two specs depend on the tool. ci::affected and the T27F dependencies view both handle the cycle, but any consumer that needs a topological order (gen order, seal closure order, a build plan) has to break it somewhere.

Boundary

• In scope: specs/igla/race/backend.t27 and specs/igla/race/rtl.t27, plus a new shared spec if one is needed. Seals and ledger rows for those files are in scope too.
• Out of scope: other igla specs, and any compiler change.

User Scenarios

• A reader opens either spec in the T27F dependencies view and sees cycle=0.
• tsort over the whole edge list prints nothing on stderr.

Requirements

• Find which names each spec actually takes from the other. rtl.t27 exports RtlModule, Signal, Assignment and Instance, among others. backend.t27 exports contains_multiply, replace_multiply, booth_encode, r_si_1_pass, and more.
• Move the shared names into one direction of the dependency, or into a new spec both can import. Behaviour must not change: every existing test in both specs keeps passing.
• Re-seal both specs and keep their ledger rows accurate.

Success Criteria

• t27c use-edges --specs-dir specs | tsort prints no loop on stderr, checked on the Railway lab.
• t27c test-report passes on both specs, and on any new spec, on the lab.
• The PR quotes the before and after edge lines.

Found by specs/t27f/spec_deps.t27. Epic gHashTag/trinity#1648.