Fix/unit scaling review followups

[novykh]

No description provided.

[ktsaou]

The performance direction is sound, particularly removing the popover remount and bounding dense canvas work. Two blockers remain:

1. The dense stacked-area reduction chooses retained samples independently per series, so adjacent layers can render different approximations of the same shared boundary.
2. The value/unit grid change is unbounded in narrow columns and the branch currently fails its own test suite.

Local validation:

• 139/140 test suites passed
• 1,345 tests passed, 2 failed, 2 skipped
• CJS/ES6 build passed
• Storybook build passed
• changed-file ESLint passed

Please address the two inline findings before merge.

[ktsaou]

The shared-index change fixes the original stacked-boundary mismatch, and the updated value/unit tests now pass. One new production correctness blocker remains in the replacement reducer: it buckets real Dygraph points by normalized x coordinates rather than canvas pixels, collapsing dense charts to only a handful of samples.

Validation on 0e59f98:

• 140/140 test suites passed (1,349 passed, 2 skipped)
• CJS and ES6 builds passed
• Storybook build passed
• changed-file ESLint passed
• local Codacy found no introduced actionable issue

[ktsaou]

No remaining blocking findings. The latest fix uses Dygraphs canvas pixel coordinates while retaining a fallback for synthetic points, and the production-shape regression test verifies every canvas bucket is represented.

Validation on 451aff0:

• 140/140 test suites passed (1,350 passed, 2 skipped)
• focused stacked-area regression test passed
• 86,400-point reproduction selected 4,806 points across 801 pixel buckets
• CJS and ES6 builds passed
• Storybook build passed
• changed-file ESLint passed
• local Codacy found no introduced actionable issue