fix: restore PDF coordinate scaling to match axes/ticks (fixes #985)

[krystophny]

Fixes #985: CRITICAL REGRESSION: PDF output plots wrong scale.

Summary

• Restore independent X/Y scaling for PDF coordinate transforms.
• Remove aspect-ratio preservation that shrank/centered content and misaligned data vs. axes.
• Aligns plot coordinate mapping with axes/tick generation which already assume full plot-area mapping.

Root Cause

• Recent refactor introduced a common-scale (min(x_scale, y_scale)) transform in fortplot_pdf_coordinate.f90.
• Axes/ticks in fortplot_pdf_axes.f90 compute positions using plot-area width/height (independent per axis).
• The mismatch caused data to be padded/centered while axes spanned the full area → perceived wrong scale.

Changes

• normalize_to_pdf_coords: use independent x_scale/y_scale; map directly to plot_area%left/bottom.
• safe_coordinate_transform: same fix; keep degenerate-range centering behavior.
• No public API changes.

Validation

• Ran make test-ci (all CI-essential tests passed).
• Manually verified PDF generation in example/pdf_coordinate_validation.f90.
• Checked backend switching tests: PDF <-> PNG switching remains OK.

Risk / Compatibility

• Restores pre-regression behavior (2+ weeks ago), matching PNG and expectations.
• Vector output fills the full plot area; no more central padding.

Follow-ups

• Consider adding a dedicated regression test that renders a small set of points and asserts PDF coordinates match expected mapping for a known plot area.