config: ignoreSelectorsVisual and toleratedDomDrift are documented but never read

[JonasJesus42]

Two .parityignore options are declared in the schema, documented in docs/config.md, and read by
nothing:

option	Status
ignoreSelectorsVisual	declared + documented, never consumed
toleratedDomDrift	declared, never consumed

For contrast, ignoreMetaKeys, ignoreConsolePatterns and ignoreRequestPatterns are all read
(checks/meta-seo.ts, checks/console-errors.ts, checks/network-summary.ts → diff/network.ts).

$ grep -rn "ignoreSelectorsVisual" packages/parity/src
src/types/schema.ts:793:  ignoreSelectorsVisual: z.array(z.string()).default([]),
src/commands/cache.ts:135:      ignoreSelectorsVisual: [],
src/commands/vitals.ts:197:      ignoreSelectorsVisual: [],

Only the declaration and two empty initializers.

Why it matters

docs/config.md presents ignoreSelectorsVisual as the way to keep a rotating banner or a
third-party widget out of the visual diff. A user who sets it gets silence, not suppression —
the noisy region keeps producing findings and there is no error, no warning, nothing to suggest the
config was ignored. That is worse than not offering the option: it costs the user a debugging
session to discover the feature does not exist.

Found while implementing #296, where I had cited it (wrongly) as the existing mechanism.

Options

1. Implement it. In checks/visual-regression.ts, drop matching regions from the DOM snapshot
   comparison and from the LLM's section list before diffing. Non-trivial: the pixel diff is
   whole-screenshot, so a selector-based exclusion means either masking the region in the bitmap or
   only filtering the semantic/DOM layers. Worth deciding which layers it is honestly able to
   cover, and documenting that limit.
2. Remove both and the docs. Least code, and expectedDivergences (score: an upgrade component should not count as a divergence forever #296) now covers the
   "expected difference" case better anyway, by reclassifying instead of blinding.
3. Warn when set but unimplemented. A one-line stopgap that at least stops the silent no-op.

I lean toward 2 for ignoreSelectorsVisual — it overlaps with expectedDivergences and the pixel
layer cannot honour it cleanly — and toward 1 or 2 for toleratedDomDrift depending on whether the
DOM-drift check still wants a per-key tolerance.

Either way it should not stay as it is.