fix(pyplot): close the eight parity gaps from the PR #44 review

[FarhanAliRaza]

Follow-up to #44, addressing all eight findings from its adversarial review. Every fix was verified against Matplotlib 3.11.0, and tests/pyplot/test_adversarial_parity_fixes.py pins the review's reproductions (plus a few neighbors: adjust-before-create, bymonth + interval striding, set_position vs subplots_adjust precedence).

Fixes

1. P1 — MaxNLocator far-offset ranges (_ticker.py): removed the dv/maxabsv < 1e-12 early return in _scale_range; only a true zero span bails now. MaxNLocator(4).tick_values(1e9+1e-4, 1e9+2e-4) returns matplotlib's exact five ticks instead of two out-of-view ones.
2. P1 — titled absolute panels (_mplfig.py): _charts builds titled panels 26/30 px taller and _panel_positions grows the placement upward to match, so the title lives in panel chrome and the plot box stays on its requested GridSpec rect.
3. P1 — dataless-axis fallback (_axes.py): the (0, 1) empty view is applied to the render snapshot (x_props/y_props) only, never persisted into axis state; render-then-plot autoscales like matplotlib.
4. P1 — GridSpec spans after subplots_adjust (_mplfig.py): span axes retain their originating SubplotSpec and are re-resolved by subplots_adjust; cell_rect falls back per-key to the figure's adjusted SubplotParams. One deviation from the review's suggested fix: 3.11 overrides a prior set_position on subplotspec-backed axes during subplots_adjust, so the spec is kept rather than cleared (test included).
5. P2 — calendar interval phase (dates.py): interval filtering is anchored at matplotlib's view-derived dtstart (vmin − relativedelta(vmax, vmin), reimplemented without dateutil); month occurrence numbers stride all months so bymonth acts as an rrule filter.
6. P2 — third-party locators (_axes.py): _nbins_hint is only written to xy Locator instances; slotted/frozen third-party locators render.
7. P2 — annotation arrows (_axes.py): callout conversion gates on arrowprops is not None and detects offset placement from the entry itself, so empty labels and arrowprops={} keep their arrows (SVG exporter verified).
8. P2 — tuple facecolors (_mplfig.py): figure facecolor normalizes through resolve_color at construction and in set_facecolor before any HTML escaping.

Verification

• Full suite: 1,825 passed (17 new regression tests).
• pre-commit run --all-files, ruff check, ruff format --check, ty check python: all green.
• Each of the eight reproductions from the review re-run against this branch and compared with Matplotlib 3.11.0 output.

[codspeed-hq]

Merging this PR will not alter performance

✅ 80 untouched benchmarks
⏩ 1 skipped benchmark¹

---

_{Comparing pyplot-review-parity-fixes (0d28e0d) with main (a7dc00c)}

Footnotes

1. 1 benchmark was skipped, so the baseline result was used instead. If it was deleted from the codebase, click here and archive it to remove it from the performance reports. ↩

[FarhanAliRaza]

Added a second commit (187308a) fixing the annotation rendering defects found in the PDSH 04.09 notebook comparison:

1. Arrows struck through their labels — arrows started exactly at the text anchor. The shared arrow geometry (_arrowgeom.py + 51_annotations.js) now takes a label_clear rectangle (the label's estimated extents) and trims the arrow start to where the departure tangent exits it, mirroring matplotlib's text-patch clipping with its visible ~0.35em + 2pt gap.
2. Client labels sat one line below the other renderers — the DOM client top-anchored annotation labels while SVG/raster (and matplotlib) anchor the text baseline; the client default is now baseline.
3. bbox'd labels anchored by box edge instead of text edge — the padded box now grows outward around the anchored text, restoring matplotlib's clearance (fixes the Thanksgiving/Christmas overlap).

Verified pixel-side against matplotlib 3.11 via headless-Chromium renders of the live client plus the SVG and native-PNG exporters. Full suite now 1,836 passing.