fix(chat): stream English answers sentence by sentence

[songoow]

Goal And Delivered Outcome

• Outcome basis / optional anchor: reproduced defect (no issue).
• Goal/source and gap: VisibleResponseStreamFilter (loopx/chat.py) holds visible answer text until a flush boundary so that a local path is never split before redaction. The boundaries were a newline and the Chinese sentence marks 。！？ only. An English answer therefore surfaced only at a newline or after 160 characters; a short multi-sentence English answer arrived all at once at the end of the Turn.
• Observable before → after: a stub Codex app-server streaming three English sentences character by character over ~7.5 s through a real loopx chat service: before one answer.delta at 7.64 s with all three sentences; after deltas at 2.06 s, 5.28 s and 7.65 s, one per sentence.
• Issue/task and intended base: none; base main.

Scope And Continuation

• Completed scope and remaining work: complete within this scope. ., ! or ? now also end a sentence when followed by a space or tab. Decimals, versions, file names and URLs (no whitespace after the point) never split. The split lands on whitespace, which the existing length fallback already treats as a safe split point without a path check, so redaction safety is unchanged.
• Slice boundary / successor: none.

Validation

• Tested revision: dd25609
• Run state: finished
• Input classes: synthetic

Check kind	Result	Public-safe evidence / limitation
integration	passed	python examples/loopx-chat-agent-smoke.py: new cases — a version point (1.2) does not split, a trailing . waits for its whitespace, each English sentence surfaces once complete; a protected path inside an English sentence streamed in 7-character chunks redacts exactly like redact_local_paths on the unsplit text. Existing long-English, long-Chinese and protected-path cases still pass.
regression_parity	passed	Failing-before check: with the source change reverted the new case sees no early output (['', '', '', '']).
unit	passed	pytest tests/test_chat_event_retention.py tests/test_chat_agent.py tests/test_chat_dsh_adapter.py tests/test_chat_codex_goal.py: 103 passed, identical to main.
real_entrypoint	passed	loopx chat from this branch and from main, each on an isolated synthetic registry with the same stub Codex app-server; SSE answer.delta timings as above.

• Coverage and gaps: the only changed path is the boundary scan in _accept_visible; covered by the filter cases and the real stream. Abbreviations such as e.g. now flush slightly earlier, which only moves a chunk boundary.

Frontend / Visual Evidence

• UI impact: none (the dashboard already renders every answer.delta; English answers now arrive earlier).

Type of Change

• Bug fix
• Test update

LoopX Area

• Host or runtime integration

Technical Direction

• Direction / acceptance reference, when applicable: Operator surface and IM integration.

Shared-authority RFC fixture impact

N/A

Boundary Checklist

• Neither the diff nor this PR body/comments/attachments disclose private state, credentials, raw traces or verifier output, internal links, or local machine paths.
• I did not duplicate maintainer-owned benchmark work unless a maintainer split out a public issue for it.
• I kept the change scoped to the linked issue/task.
• I completed the visual evidence section for UI changes, or marked UI impact none.
• Every commit includes a DCO Signed-off-by trailer (git commit -s).

Future-facing refactor pass: considered making the boundary set configurable per locale; unnecessary, since the rule is locale-neutral punctuation plus whitespace.