Splanc family prototypes and native PnR exploration

[fughilli]

This checkpoint brings the Splanc family prototypes and the current native placement/routing work into one reviewable source tree. Mini placement now explores multiple legal layouts before detailed routing, including bottom-side pogo locations beneath suitable opposite-side bodies. The board is still incomplete; this remains a draft stacked on pnr-system / PR #126.

Collaborator starting point: engineering checkpoint and navigation guide.

Included work

• Native routing and placement experiments: multilayer capacity scoring, coordinated pad access, incremental copper preservation, power/plane and USB handling, via/track consolidation, and electrical/DRC acceptance guards. Some improvements remain in isolated experimental source overlays; the guide distinguishes these from the production build.
• An eight-design native regression ladder, profiling and Rust search support, and a live viewer with phase inspection, immutable annotated snapshots and persistent worker controls.
• Splanc / Splanc GNSS and MAX electronics prototypes, BOM/pricing models, and parametric enclosure, section-viewer and Blender rendering scripts.

Validation and current limits

• Native regression ladder: 16/16 saved boards passed, with zero KiCad unconnected items and zero DRC findings, plus original pin/net preservation, source-width and pad-entry checks. Fixtures range from a two-component connector/LED board (external current-limited supply) to a 20-component TLC555/CD4017B LED chaser and a four-layer plane variant. This is backend integration coverage, not complete atopile or analog-circuit qualification.
• Two representative initial-placement-pool cases were rerun after the latest hard-side/under-body changes and passed the same native gates. Their 100 layer-PDF pages and all four 5 mm via clusters were visually reviewed; remaining route-quality opportunities are recorded.
• Placement verification: 67 tests passed; the 12 initial-pool tests passed again after live phase-label changes. Changed Python sources parse, the viewer JavaScript passes syntax checking, and the source diff passes whitespace checks. These are scoped checks, not a claim that every historical experiment was rerun.
• Protected Mini best: 48 native opens, zero native violations. It is neither fully routed nor electrically qualified. A separate 57-open pogo trial failed preservation/electrical gates and was not promoted.
• Fresh119 is running the actual atopile-to-native-board target with the initial pool enabled. No success or convergence is claimed for that unfinished run. Budget exhaustion is distinguished from a measured plateau.
• Family boards and enclosures remain engineering prototypes, not manufacturing/tooling releases.

Source and artifact policy

This is a normal fast-forward from the previous PR head. The full unpublished local commit series is preserved locally, while the published checkpoint avoids introducing its historical release ZIPs. New Git blobs total approximately 8.1 MB; no release ZIPs, generated videos, installed runtimes or output directories are added.

Artifact inventory records omitted assets and hashes. Generated boards, PDFs, renders and model bundles remain local. The guide documents reproduction commands and limitations: a fresh clone alone does not yet reproduce every product render because some reference/model inputs are local derived assets.