Fix hibernated transcript previews

[gbasin]

Summary

• Tail-read large transcript logs so hibernated session previews return quickly.
• Add byte-cursor pagination so older entries load from the end without rescanning full logs.
• Harden tail cursors for CRLF logs and multibyte UTF-8 split across read chunks.
• Bump package version to 0.3.2.

Root Cause

Large transcript previews were read and scanned from the start, which could exceed the client timeout for hibernating sessions with multi-megabyte logs. The initial tail-reader also needed byte-level cursor handling to avoid corrupting split UTF-8 characters or miscounting CRLF logs.

Validation

• bun run lint && bun run typecheck && bun run test
• Served locally from this branch over Tailscale and verified the app loaded at http://100.82.11.20:5173/