Python: Normalize durable workflow inputs

[moonbox3]

Motivation & Context

Keep reserved workflow input handling consistent across durable hosts.

Description & Review Guide

• What are the major changes? Normalize reserved input markers at durable workflow boundaries and cover the behavior end to end.
• What is the impact of these changes? Ordinary JSON and trusted child-workflow inputs retain their existing behavior.
• What do you want reviewers to focus on? Boundary handling and compatibility coverage.

Related Issue

N/A

Contribution Checklist

• The code builds clean without any errors or warnings
• All unit tests pass, and I have added new tests where possible
• The PR follows the Contribution Guidelines
• This PR is linked to an issue and there is no other open PR for this issue (see Related Issue above).
• This is not a breaking change. If it is a breaking change, add the breaking change label (or add "[BREAKING]" to the title prefix, before or after any language prefix) — a workflow keeps the label and the title prefix in sync automatically.

[Copilot]

Pull request overview

This PR normalizes handling of framework-reserved durable-workflow input markers (e.g., __pickled__, __type__) at durable host boundaries so untrusted “marker-shaped” inputs are consistently neutralized before reaching workflow executors, aligning behavior across Durable Task and Azure Functions hosts.

Changes:

• Durable Task orchestrator now strips pickle/type markers from initial input even when the start executor’s input type can’t be selected (e.g., union/unannotated cases).
• Azure Functions workflow /run route now strips pickle/type markers in addition to sub-workflow envelope markers before scheduling the orchestration.
• Adds end-to-end behavioral tests for both hosts validating neutralization of reserved marker-shaped initial input while preserving ordinary JSON behavior.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.

File	Description
python/packages/durabletask/tests/test_durabletask_workflow_initial_input.py	New behavioral tests validating marker-shaped initial inputs are neutralized and regular union-shaped inputs are preserved.
python/packages/durabletask/agent_framework_durabletask/_workflows/orchestrator.py	Ensures initial input is sanitized via strip_pickle_markers() even when no primary input type is selected.
python/packages/azurefunctions/tests/test_azurefunctions_workflow_initial_input.py	New behavioral test validating the Azure Functions /run route neutralizes reserved marker-shaped input.
python/packages/azurefunctions/agent_framework_azurefunctions/_app.py	Sanitizes untrusted HTTP client input with strip_pickle_markers() before start_new().

[github-actions]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/azurefunctions/agent_framework_azurefunctions
_app.py	585	130	77%	82–83, 86–89, 317, 375, 435–437, 439, 459, 463, 466, 468, 470, 494, 497–500, 534–536, 538, 544–545, 552, 554, 569–574, 587, 589, 604–605, 607–608, 613–615, 617–620, 624, 629–632, 636, 642, 695, 732, 738, 741, 744, 889–890, 998, 1006–1007, 1027–1029, 1035–1037, 1043–1045, 1078–1079, 1139–1140, 1189–1190, 1195, 1277, 1280, 1289–1291, 1293–1295, 1297, 1299, 1310–1313, 1315, 1317–1318, 1320, 1327–1328, 1330–1331, 1333–1334, 1336, 1340, 1350–1352, 1354–1355, 1357–1359, 1366, 1368–1369, 1371, 1392, 1397, 1409, 1481, 1571, 1586–1589, 1614
packages/durabletask/agent_framework_durabletask/_workflows
orchestrator.py	457	157	65%	209–211, 263–265, 341–342, 344–351, 353, 360, 378–386, 492, 495–497, 500–502, 505, 507–512, 514–517, 519, 521–525, 538, 540–541, 543–548, 550–551, 553–554, 556–557, 559, 573–576, 583, 596, 603–605, 607–608, 610, 628–629, 745, 787, 793, 818, 822, 828–835, 837–838, 897, 944–945, 947–950, 959, 1059–1061, 1068–1074, 1084, 1090, 1092, 1099, 1122, 1124, 1137, 1144–1145, 1147–1148, 1152–1153, 1161–1164, 1166–1168, 1191, 1193, 1207, 1213–1214, 1216–1217, 1224–1229, 1235–1237, 1242, 1244–1245, 1250, 1252, 1259, 1264
TOTAL	44684	5102	88%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
9083	33 💤	0 ❌	0 🔥	2m 22s ⏱️

[github-actions]

Automated Code Review

Reviewers: 5 | Confidence: 91% | Result: All clear

Reviewed: Correctness, Security Reliability, Test Coverage, Failure Modes, Design Approach

---

Automated review by moonbox3's agents