[imakris]

Why

A horizontal-axis cache miss creates and destroys its tick-candidate vector for each density level. Preparing a built-in line or area also copies the newly computed segment-span vector into the draw-command handoff even though the preparation code can read the transferred storage.

Change

• Keep the candidate vector alive for one axis calculation, clear it at each density level, and reuse its capacity. Reserve remains inside the existing step-preparation scope so early-rejected levels do not cause a new allocation.
• Move computed segment spans into the caller's output vector, then use a const reference to that storage for the remainder of preparation. With no output vector, preparation continues to read the local storage. Existing early-return behavior and recorded span contents are preserved.

These are local ownership and allocation changes; tick selection, formatting, sample segmentation, and draw ordering remain the same. The likely benefits are axis-cache misses and discontinuous line/area frames. End-to-end gains have not been measured.

Validation

• Baseline source bytes were verified against the pinned Git tree; both diffs pass whitespace checks.
• Inspected the full preparation lifetime, including early success returns and the later draw-command move. Inspected existing LayoutCalculator and SampleUploadReuse test coverage.
• Native tests were not run locally because CMake and Qt 6.12 are unavailable. Public QRhi and QRhi+Text PR workflows can provide build/test evidence; their results must be reviewed separately.

Before marking ready

Run LayoutCalculator and SampleUploadReuse in the supported QRhi configurations, and measure an axis-cache-miss/segmented-series workload against the baseline. This PR is a draft pending native and performance evidence.

Independent review

Fresh-context independent source review found no remaining code blocker in this exact candidate. Native tests and representative application performance evidence remain pending as described above.