grt: enhance CUGR's congestion handling

[eder-matheus]

Summary

Adds CUGR's missing stage-4 iterative rip-up and re-route (RRR) loop, plus the supporting infrastructure to fight per-layer overflow concentration on heavily-adjusted designs.

Changes on src/grt/src/cugr/:

• New stage 4 (CUGR::iterativeRRR) wraps the maze stage in a loop that sharpens the logistic cost slope per iteration (via the new GridGraph::setCostMultiplier) and widens the rip-up set to nets touching near-overflow edges (via the new GridGraph::checkCongestion(tree, threshold) + CUGR::updateCongestedNets). Emits GRT-0117 per iteration and GRT-0118 if overflow remains, mirroring FastRoute's GRT-0102 style.
• Iteration budget honors the existing -congestion_iterations flag; when -use_cugr is set with no explicit value the default is 5 (FastRoute keeps its default of 50).
• GridGraph adjustment loop now mirrors FastRoute's floor-to-1 rule (GlobalRouter.cpp:1812): edges with any pre-adjustment capacity keep at least 1 usable track unless the user explicitly asked for 100% reduction.
• CUGR::saveCongestion rounds per-edge overflow with ceil(demand - cap) so the marker tree still gets entries when overflow is fractional.
• CUGR::debugCongestion2D (gated on set_debug_level GRT rrr_2d 1) reports the gap between 3D overflow and 2D-aggregate overflow — the spreadable portion of the residual.
• PatternRoute::calculateRoutingCosts reads the per-net layer range (GRNet::getLayerRange) and branches fixed_layers.isValid() so non-pin Steiner nodes don't get clamped to an inverted interval.

Type of Change

• New feature
• Bug fix

Impact

For the canonical stress test (cugr_adjustment3, gcd_nangate45 + 90% layer adjustment): total overflow drops from 445 → 18 units (-96%); wirelength is essentially unchanged (+0.3%); via count rises 36% (4225 vs 3095) as the router accepts longer paths and more vias to absorb redistributed traffic. Other CUGR designs that were already clean remain so. RRR is a no-op when totalOverflow() == 0, so default-flow designs incur only the gate check.

-congestion_iterations N now also drives CUGR (it was FastRoute-only before).

12 .ok / .guideok baselines on the CUGR test suite refreshed to capture the new routings.

Verification

• I have verified that the local build succeeds (./etc/Build.sh).
• I have run the relevant tests and they pass.
• My code follows the repository's formatting guidelines.
• I have included tests to prevent regressions — the existing CUGR tests (cugr_adjustment1-3, critical_nets_percentage_cugr, clock_route_cugr*, etc.) now exercise the RRR loop and the adjustment-floor path; their baselines are updated to lock the new behaviour.
• I have signed my commits (DCO).

Related Issues

None.

[gemini-code-assist]

Code Review

This pull request introduces an iterative Rip-up and Re-route (RRR) mechanism to the CUGR global router, aimed at resolving persistent congestion by sharpening cost gradients and widening the rip-up set across multiple iterations. Key additions include a 2D congestion diagnosis tool, a configurable iteration limit, and a 'floor-to-1' capacity adjustment rule to ensure at least one usable track remains on adjusted layers. Feedback identifies a critical bug in the congestion reporting logic where the removal of a guard clause causes all edges to report overflow regardless of actual demand. Additionally, it is recommended to refine the early-exit condition in the RRR loop to consistently ignore fractional overflows below 1.0.

[github-actions]

clang-tidy review says "All clean, LGTM! 👍"

[github-actions]

clang-tidy review says "All clean, LGTM! 👍"