Python: Fix DevUI streaming memory growth regression

[eavanvalkenburg]

Motivation and Context

Fixes #5877.

DevUI streaming can retain too much browser-side state during long streamed answers, causing renderer memory growth. This change bounds the main retained state paths while preserving streaming and refresh-resume behavior with a capped preview.

Description

• Bound persisted streaming resume text to a capped preview instead of storing the full accumulated response in localStorage.
• Cap retained debug events and truncate oversized text deltas before storing them in the debug panel state.
• Share memoized processed debug events between debug tabs/counts to avoid repeated full-history processing during streaming.
• Strengthen the browser memory regression test with probes for streaming-state storage size and retained debug DOM event count.
• Regenerate the shipped DevUI frontend bundle.

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the Contribution Guidelines
• All unit tests pass, and I have added new tests where possible
• Is this a breaking change? If yes, add "[BREAKING]" prefix to the title of the PR.

[Copilot]

Pull request overview

This PR addresses a DevUI browser-side memory growth regression during long streaming responses by bounding persisted streaming-resume state, limiting retained debug history, reducing repeated processing work in the debug panel, and strengthening the browser-based regression test to detect these issues.

Changes:

• Cap persisted streaming resume text to a bounded tail preview (with an explicit “preview” marker) instead of storing the full accumulated response.
• Limit retained debug events and truncate large text deltas before they enter the debug panel state; memoize processed debug events to avoid repeated full-history processing during streaming.
• Expand the DevUI streaming memory regression test to probe localStorage streaming-state footprint and retained debug DOM node counts.

Reviewed changes

Copilot reviewed 7 out of 8 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
python/packages/devui/tests/devui/test_ui_memory_regression.py	Adds browser probes for localStorage streaming-state size and debug event DOM retention, and asserts new caps during streaming.
python/packages/devui/frontend/src/stores/devuiStore.ts	Caps retained debug events and truncates oversized text deltas before storing debug events.
python/packages/devui/frontend/src/services/streaming-state.ts	Introduces bounded accumulatedText preview behavior and records whether the stored text is a preview.
python/packages/devui/frontend/src/services/api.ts	Preserves the new accumulatedText preview flag when resuming streaming state.
python/packages/devui/frontend/src/components/layout/debug-panel.tsx	Adds DOM markers for debug event items and shares memoized processed events across tabs/counts to reduce reprocessing.
python/packages/devui/frontend/src/components/features/workflow/workflow-view.tsx	Refactors workflow event timestamp extraction into a shared helper.
python/packages/devui/frontend/src/components/features/agent/agent-view.tsx	Restores streaming text with a prefix when only a preview tail is available after refresh.

[moonbox3]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
TOTAL	36298	4308	88%

report-only-changed-files is enabled. No files were changed during this commit :)

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
7244	34 💤	0 ❌	0 🔥	1m 59s ⏱️

[github-actions]

Automated Code Review

Reviewers: 4 | Confidence: 88%

✓ Correctness

No actionable issues found in this dimension.

✓ Security Reliability

No actionable issues found in this dimension.

✓ Test Coverage

The PR adds browser-level memory probes to the existing E2E regression test, validating that streaming state storage and debug event DOM count stay within bounds. The assertions are structurally sound (debug panel renders without virtualization and is open by default). However, the new assertions only check upper bounds and could pass vacuously if the features under test fail to render or persist data at all. A lower-bound assertion would strengthen confidence that the caps are actually being exercised. No frontend unit test infrastructure exists in the project, so the absence of unit tests for prepareDebugEvent, the streaming text cap logic, and getWorkflowEventTimestamp is a pre-existing gap rather than a regression introduced by this PR.

✗ Design Approach

That silently presents truncated preview text as if it were the full streamed answer in that UI flow. The capped preview approach fixes the storage growth, but it also introduces one functional regression in the refresh-resume path: after the preview limit is hit, a page refresh resumes from only the stored tail, so the final in-progress assistant message silently loses the earlier prefix of the response.

---

Automated review by eavanvalkenburg's agents