fix: preserve UTF-8 characters split across PTY reads

[dgokeeffe]

Problem

CoDA decodes each arbitrary PTY read independently with replacement enabled. When a multibyte UTF-8 character is split across reads, the partial sequences are permanently converted to replacement characters. Large Pi and OpenCode outputs make this readily observable through an Omnigent-attached host.

Acceptance criteria

• Decode PTY output incrementally across read boundaries.
• Flush decoder state when the stream closes.
• Ensure spawned terminal sessions use a UTF-8 locale.
• Add regression coverage for split multibyte characters.
• Verify a 200-line mixed-Unicode response through Omnigent attached to CoDA with zero replacement characters.