Live e2e: documents connector object-store support on Azure Blob and GCS

[abbccdda]

Context

Follow-up from PR #171 review comment: #171 (comment)

PR #171 added S3 / object-store support to the documents connector (source path and image_store can be s3:// prefixes) and validated it with a live e2e run against real S3 (us-east-1). The object_store crate backing this also supports Azure Blob Storage and GCS, so the same code path should work against those providers — but it hasn't been exercised live.

Ask

Run the same live e2e coverage PR #171 did on S3, but against:

• Azure Blob Storage — we have Azure credits.
• GCP Cloud Storage — only if we already have GCP credits; otherwise skip/defer.

Scope

Mirror the S3 e2e checklist from the PR against each provider:

• Nested corpus scan with glob + prefix scoping (sibling-prefix decoy and non-matching extensions excluded).
• image_store page-render / crop uploads land as real objects with the correct Content-Type, and the write-probe object is cleaned up.
• Self-ingestion guard holds when the image_store prefix sits inside the source path.
• Negative startup cases fail with actionable errors: cross-container/cross-bucket image_store, nonexistent container/bucket, credentials smuggled into options, missing region/credentials in env, and a write preflight failure on a read-only target.

Notes / open questions

• Confirm the env-only credential contract and region/endpoint config translate cleanly to Azure (AZURE_STORAGE_ACCOUNT / AZURE_STORAGE_ACCESS_KEY etc.) and GCS (GOOGLE_SERVICE_ACCOUNT / ADC), and update docs/documents.md if the per-provider setup differs.
• The same-bucket constraint may need a per-provider analog (same container / same bucket).