fix(bedrock): guard against null url in image_url parts and lifecycle status (#55686)

[AlexFucuson9]

Summary

Prevent AttributeError crash when Bedrock Converse encounters an image_url part with a null url field, or a model lifecycle entry with a null status.

Problem

Two .get(key, "") patterns crash when the key exists with value None:

1. image_url.get("url", "") returns None (not "") when url is explicitly null → url.startswith("data:") raises AttributeError: 'NoneType' object has no attribute 'startswith'

2. lifecycle.get("status", "").upper() returns None when status is null → .upper() raises AttributeError

The dict.get(key, default) default only applies when the key is absent, not when it is present with value None.

Fix

Use (x.get("key") or "") which coalesces both absent-key and None-value to empty string:

• image_url.get("url") or "" — null URL parts are safely skipped (empty string doesn't match data: prefix)
• lifecycle.get("status") or "" — null status is treated as non-ACTIVE (model skipped)

Fixes #55686

[alt-glitch]

This was generated by AI during triage.

Related: this fixes the same image_url null-url crash (#55686) as the earlier open PRs #55685 (earliest, has regression tests) and #55699, and additionally guards a null modelLifecycle.status in discover_bedrock_models. It's a superset of the null-url fix, not a duplicate. Maintainer to pick the canonical PR for the shared image_url fix (#55685 looks canonical: earliest + regression tests) and decide whether to fold in this PR's extra lifecycle-status guard.

[teknium1]

Thanks for the focused Bedrock hardening. The stated null cases remain reproducible from current main: agent/bedrock_adapter.py:522-523 can call startswith on a null URL, and agent/bedrock_adapter.py:1147-1148 can call upper on a null lifecycle status.

Problems

• The PR changes only agent/bedrock_adapter.py; it adds no regression coverage. Existing tests cover a valid image data URL at tests/agent/test_bedrock_adapter.py:332-348 and ACTIVE/LEGACY lifecycle states at :711-773, not either null case.
• The proposed or "" expressions fix None, but truthy non-string values can still reach startswith or upper. The analogous Gemini converter uses an explicit string guard at agent/gemini_native_adapter.py:213-215.

Suggested changes

• Add direct null-URL and null-lifecycle-status regression tests in tests/agent/test_bedrock_adapter.py.
• Use explicit string normalization for both values before their string-method calls.

Automated hermes-sweeper review.

[teknium1]

This fixes None, but a truthy non-string value such as 123 still reaches url.startswith(...) below. Please normalize with an explicit isinstance(url, str) guard, matching the defensive Gemini converter.

[teknium1]

This coalesces null status values, but truthy non-strings still reach .upper(). Normalize to a string before comparison so malformed lifecycle payloads cannot retain the same crash class.