v0.5.0 — Read-aloud speed-up & macOS visual refresh

Read-aloud — faster & more robust

• Text cleanup (#33): collapses double spaces / tabs / newlines from terminal copies before synthesis.
• Sentence batching (#33): long passages split on ./!/? into ≤400-char batches — first audio starts quickly, and a huge blob never hits the provider as one slow, failure-prone request.
• Lazy prefetch (#34/#36, wired by #37): cloud backends (OpenAI, Deepgram) synthesize the next batch(es) while the current one plays — no inter-batch silence — and stop fetching the instant the listener stops early. Additive TtsProvider API (can_prefetch/prepare/play_prepared); the native voice keeps the sequential path.

Visuals — modern macOS look

• Embedded Inter font (#35): replaces the blocky 5×7 all-caps bitmap with proportional, anti-aliased, mixed-case text across the recording pill, read-aloud popup, and clipboard toast.
• macOS system-color palette (#38): dark-mode popover material + Apple system colors, unified across the three overlays.

Quality

Zero build warnings, 101 tests passing, clippy clean.

Full changelog: v0.4.0...v0.5.0

What's Changed

• feat(app): normalize and batch text for read-aloud by @joeVenner in #33
• Feature/read aloud prefetch by @joeVenner in #36
• feat(app): modern macOS overlay text via embedded Inter font by @joeVenner in #35
• fix(app): wire up read-aloud prefetch dropped in merge by @joeVenner in #37
• style(app): macOS system-color palette across the overlays by @joeVenner in #38
• chore(release): v0.5.0 by @joeVenner in #39

Full Changelog: v0.4.0...v0.5.0