Provide a uniform serialization contract (to_dict/to_json) across trigger binding types

[larohra]

Summary

Trigger/input binding objects lack a consistent, machine-usable serialization contract. Some expose to_dict(), some expose to_json() (returning a str), and most expose neither. Any consumer that needs to turn a binding object into structured data (logging, diagnostics, message pipelines, or feeding an LLM prompt) must special-case every type, and the natural fallback — str(obj) — yields an unusable Python repr (<azure.functions.blob.InputStream object at 0x...>) that silently drops the entire payload.

The inconsistency today (azure-functions 2.2.0)

Binding type	to_dict()	to_json()	get_body()	get_json()	str(obj) result
blob.InputStream	❌	❌	❌ (read())	❌	<...InputStream object at 0x...>
QueueMessage	❌	❌	✅	✅	<azure.QueueMessage id=... > (body lost)
ServiceBusMessage	❌	❌	✅	❌	<...ServiceBusMessage object at 0x...>
EventGridEvent	❌	❌	❌	✅	<azure.EventGridEvent id=.. topic=.. > (data lost)
EventHubEvent	❌	❌	✅	❌	<azure.EventHubEvent partition_key=.. > (body lost)
KafkaEvent	❌	❌	✅	❌	<azure.KafkaEvent key=.. > (body lost)
Document	✅	✅	—	—	(works)
DocumentList	❌	❌	—	—	[<azure.Document ...>]
SqlRow / MySqlRow	❌	✅ (str)	—	—	<SqlRow at 0x...>
SqlRowList / MySqlRowList	❌	❌	—	—	[<SqlRow ...>]
TimerRequest	❌	❌	—	—	<...TimerRequest object at 0x...>

So four different conventions (or none) for the same conceptual operation: "give me this binding as JSON-safe data."

Impact

Downstream consumers must maintain a per-type adapter matrix and re-derive field names (name/uri/metadata, get_body(), get_json(), enqueued_time, partition_key, …), including bytes→text/base64 and datetime→ISO handling, purely to avoid the str() repr trap. This is fragile across SDK versions and duplicated in every project that needs it.

Proposal

Introduce a uniform, documented serialization contract implemented consistently across all binding types, e.g. one of:

1. A to_dict() -> dict (JSON-safe) on every binding object, with a companion to_json() -> str, consistent in name and return type (today to_json returns a str on rows but a JSON string elsewhere, and to_dict exists only on Document); or
2. A shared Serializable/SupportsToJson protocol (or a mixin base) that guarantees the method surface and centralizes bytes/datetime normalization.

Guidelines for the payload: bytes decoded as UTF-8 when valid else base64 (with an encoding marker), datetimes as ISO-8601, and blob bindings surfaced as name/uri/length/metadata without forcing a read() of content.

Context

Found while building an LLM-agent runtime on Azure Functions, where trigger data is serialized into the agent prompt. We worked around it with a per-type adapter layer, but a first-party contract would let all consumers drop the shim. Downstream reports: Azure/azure-functions-agents-runtime#85 and Azure/azure-functions-bucees-planning#1197.