Give update handlers the committed workflow input and signal snapshot

[rmcdaniel]

Problem

Rust update handlers receive QueryContext, but Worker::execute_update_task in the published 3.3.0 source initializes workflow input to null and signal deliveries to an empty list. Queries populate those same documented context methods from the committed task snapshot.

An update that reconstructs state from original input and signals therefore sees different state from a query for the same run.

Acceptance

• Populate JSON and lossless Avro workflow input from the original task arguments, preserving separate update arguments.
• Populate committed signals with the same official decoder and ordering as queries.
• Reject malformed context payloads before invoking the update handler.
• Add focused source regressions and a retained replay fixture.
• Publish a patch and verify the exact installed crate in Sample App's live query/update snapshot check before and after worker replacement, retaining its existing nine client/handler directions and duplicate-completion checks.

Consumer qualification: durable-workflow/sample-app#155. This is a prerequisite for durable-workflow/.github#122's stateful update/query work.

[rmcdaniel]

Confirmed with the installed Rust 3.3.0 crate and published Server 2.5.9 in Sample App's live run.

Runner e512cdc46b24cced0e908f6aa6c2b540f5831e06 executed all nine client/handler directions before the new check. On the same original Rust run, the query returned its original workflow input and one committed updates-touch signal. The update completed and persisted a result with workflow_input: null and signals: []. This is a product failure. The experiment removed its stack, volumes and four task images after failing.

The focused source regression reproduces the same null input on the baseline. Both new source regressions pass with PR #74, including rejection of malformed context payloads before calling the handler. Normal source qualification is running. Next: publish the qualified patch and repeat the live check before and after worker replacement.

[rmcdaniel]

The exact published 3.3.1 consumer run is a product failure. Original workflow input is now correct, but the update snapshot contains two copies of the one acknowledged signal. The same-run query contains one. There is one SignalReceived delivery and one persisted update completion.

The full update history contains both receipt and application markers for the signal. The shared decoder counted both as deliveries. The assertion remains strict. PR #75 matches immutable signal/command identity, preserves distinct equal-valued signals and rejects conflicting data for one identity. It adds an official-worker corpus fixture and targets patch 3.3.2. #73 remains open until the installed patch passes the initial and replacement checks.

[rmcdaniel]

Fixed in published Rust SDK 3.3.3, source 13c0d50f148bb8136e7ce46c9009f68ecc34e35b. The independently fetched Cargo index checksum is 10f10a62d9691094f654557ae8c45a2abd0b608e389860b0683a559b9936912f. Publication qualification verifies the exact registry archive against source and builds a fresh Rust 1.86 consumer.

Updates now receive the original workflow input and committed signal argument lists. Receipt/application history records count once per immutable signal identity. Decoding preserves nested arrays, empty arguments and typed Avro values while accepting both supported application marker forms and rejecting conflicting history.

Sample App #155 is merged. Its main published-stack experiment passes against PHP 2.2.3, Python 2.4.3, Rust 3.3.3, CLI 2.1.3, Workflow 2.5.3, Waterline 2.3.1 and Server 2.5.9 at sha256:11c11e11498b63e644d906e43f49810a1bc0b5298f750734fe0edb279c622bca.

The same original Rust run returns its original input and exactly three signal argument lists from both query and update, before and after SIGKILL/replacement. These cover a map argument, one nested array argument and no arguments. Each signal has one applied history identity, and each snapshot result matches its one durable update completion. All nine client/handler directions, absent-worker acceptance, duplicate identity, handler failure diagnostics, validator refusal and completion of all three original runs pass. The experiment removes its stack, volumes and task images.

Supported Rust 1.86 and stable source checks pass, including 481 unit tests, Avro protocol checks and the replay regression corpus. This completes the SDK defect. The broader experiment audit continues in durable-workflow/.github#122.