Fix rangeslider "fixed" rangemode using placeholder range of autoranged counter axis - #7967

Exact-head verification review (diff walked against upstream/main ba10ef79d, PR head 11f79e4f).

Root cause - confirmed

• src/plots/cartesian/set_convert.js (~L492) falls back to constants.DFLTRANGEY = [-1, 4] when a y axis has no valid range at supply-defaults time.
• src/components/rangeslider/defaults.js (~L67-72) then defaults the counter-axis rangeslider.yaxis.range off yAxOut.range.slice() for any rangemode !== 'match' - so for fixed on an autoranged counter axis, the slider range starts life as the [-1, 4] placeholder.
• doAutoRange (src/plots/cartesian/autorange.js ~L401) only refreshes axeRangeOpts.range when rangemode === 'auto', so the stale placeholder survives to draw; the opposite-axis masks then shade everything outside the disagreement between [-1, 4] and the real data band, which is exactly the two grey bands in the screenshots.
• The toggle behavior also checks out: switching rangemode away and back forces a second supply-defaults pass, by which time the y axis holds its computed range, so the re-coerced fixed range is correct.

Fix assessment

• Flagging only defaulted-from-placeholder fixed ranges (!rangemodeDflt && yAxOut.autorange) and routing them through the existing 'auto' branch in doAutoRange is a minimal fix with no schema change. Delete-on-consume gives "resolve once, then stays put", which matches the fixed contract: later autorange changes correctly leave the slider alone (covered by the restyle step in the new test).
• Internal-key hygiene checked: the flag lives on the full-layout container, and by the time anchorAx._input.rangeslider[ax._name] = Lib.extendFlat({}, axeRangeOpts) copies state back toward user input the flag has already been deleted, so nothing leaks into gd.layout. Fresh newPlot calls start from clean containers, so no stale-flag path there either.

Nits (non-blocking)

1. The new test locks in the numeric behavior but not the internal-key claim from the PR description. Cheap insurance: assert '_rangeDfltFromAutorangedAx' in gd._fullLayout.xaxis.rangeslider.yaxis is false after the first plot (and/or absent from the layout input). That guards against someone later moving the extendFlat copy above the delete.
2. The reported repro includes the rangemode toggle clearing the bands. A follow-up Plotly.relayout(gd, 'xaxis.rangeslider.yaxis.rangemode', ...) round-trip asserting the slider range stays sane would pin down the second-supply-defaults path this PR deliberately leaves untouched.
3. Explicit-but-invalid rangeslider.yaxis.range together with fixed: rangemodeDflt stays undefined, so the flag applies and the range resolves once from autorange. I believe that is the right fallback (user asked for fixed with an unusable range), but a line in the draftlog entry would stop future readers from mistaking it for an accident.

CI note: only the Orca security scans have run on this head so far; the jasmine suites are maintainer-gated, so the new test has not executed yet. Mergeable state verified via API at review time.