[brianmk]

Expose each line segment's midpoint as a real solver vertex so it can be snapped to and constrained like any other point:

• Sketch::addLineSegment adds a free parameter for the midpoint, tied to the endpoints by an internal P2PSymmetric constraint. The constraint is solver-only (rebuilt from the geometry on load) and never serialized, so no file-format change is needed.
• Vertex mapping (SketchObjectGeometry::rebuildVertexIndex, GeoList) and the point getters expose PointPos::mid for lines.
• The Sketcher edit-mode coin converter draws start/end/mid for lines and keeps the midpoint hidden (per-vertex transparency) until it is hovered/selected.
• SketcherGui: Vulkan-only hover-midpoint marker (gated by FREECAD_USE_VULKAN).

Revert note: this adds a vertex per line, so sketch "Vertex N" sub-element numbering shifts for every line after the first (a line's mid vertex is inserted between its end and the following geometry's vertices). Documents that reference line vertices externally by number may need review.

---

Part 2 of a stacked series; base is the Gui: viewport renderer label branch.

[brianmk]

CI investigation: the Ubuntu / Build failure on this branch is caused by this PR's own commit, so I have dropped it from the stack and I'm closing this PR rather than merging it.

Symptom (identical on #143, #144, #145, #146 — all stacked on this branch):

FAIL: testNegativeCone (PartDesignTests.TestHelix.TestHelix.testNegativeCone)
  AssertionError: -0.17596393545793754 != 6.0643 within 4 places
FAIL: testPartDesignTNPChamfer  AssertionError: 44 != 62
FAIL: testPartDesignTNPFillet   AssertionError: 44 != 62
Ran 3558 tests ... FAILED (failures=3, skipped=39, expected failures=5)

Bisect (locally, against an up-to-date build):

• Gui: show the viewport renderer label from the active backend #142's tip af34c81031 passes the same 3558 tests (CI run 37736182463, Oct 8).
• This PR is af34c81031 + one commit (c061658d1c), so the regression is in it.
• Reverse-applying the App-side hunk (Sketch.cpp, GeoList.cpp, SketchGeometry.cpp, SketchObjectGeometry.cpp) makes all three tests pass, on the same build.
• Keeping the extra midpoint parameters but removing only the internal rule (GCSsys.addConstraintP2PSymmetric(p1, p2, pmid)) also makes all three pass — the failure is the solver equation, not the extra unknown.
• Keeping the vertex-index/getPoint exposure without any solver change is harmless.

Why it can't be fixed by tweaking the rule: the midpoint equations are linear and implied for any line endpoints, yet adding them changes which point DogLeg converges to on underdetermined sketches. That silently changes existing PartDesign results (helix volume, TNP element maps) and can even fail the solve (Invalid solution from DogLeg solver, Both points are equal). The commit's own revert note also flags the sub-element renumbering. A future attempt should keep the midpoint out of the solver (e.g. translate Coincident(line.mid, x) into constraints on the line's endpoints, or create the midpoint unknown lazily only when a constraint references it) rather than adding an equation to every line.

The commit is preserved at c061658d1c on branch sketcher/line-midpoint-vertex (untouched) if you want to revive it as a standalone, opt-in feature with its own tests.