fix(text): honor element color/anchor on parse_colors path and drop O(n²) fitting

[balloob]

Summary

Correctness and performance fixes for the text and multiline elements, focused on the parse_colors path and on text fitting.

Findings addressed

• A6 — parse_colors ignored element color. parse_colored_text hardcoded black for text outside any [color] markup, so {value: "plain text", color: red, parse_colors: true} rendered black. The helper now takes a default_color argument and draw_text/draw_multiline pass the element color through.
• A7 — parse_colors dropped the vertical anchor for single-line text. The single-line branch drew at the raw y with anchor="lt", so anchor: mm/mb were silently ignored and the text shifted down by half its height the moment parse_colors was enabled. The vertical anchor offset is now applied, matching the plain path.
• B2 — O(n²) truncation/wrapping. truncate: true shrank the string one character at a time, re-measuring the whole string each step (measured at ~3s for a 1600-word string — that blocks HA's event loop). It now binary-searches the cut point (O(n log n)). Word-wrap accumulates precomputed word widths instead of re-measuring the whole current line per word.
• B7 — redundant textbbox after draw.text. draw_multiline called textbbox twice and threw the result away (pure waste); those are removed. The per-segment textbbox calls on the parse_colors paths of draw_text are replaced with cheap font-metric / precomputed block-height math for the pos_y cursor.

Note / deferred

Finding A7 also mentions a related inconsistency: on the parse_colors path, calculate_segment_positions applies both align and the horizontal anchor component, so anchor: mm + align: center double-offsets horizontally. Fixing that cleanly depends on the intended semantics of align vs anchor on this path (Pillow treats align as intra-block justification only), so I've left it out of this PR to avoid guessing at behavior — happy to follow up if you confirm the desired semantics.

Tests

Adds regression tests for A6 (element color applied), A7 (vertical anchor respected), and unit tests for the B2 truncate/wrap helpers (including a check that binary-search truncation matches the old linear result). No new failures in the suite (pre-existing tests/visual/* byte-snapshot failures under Pillow 12 are unrelated).

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!