remotecache: Refactor S3 cache backend to use minio-go client

[sorenhansendk]

The MinIO SDK is designed to talk to any S3-compatible object storage, and the project actively maintains interoperability across providers. This directly addresses the signature/header edge cases we’ve hit with S3-compatible endpoints (e.g., checksum and Accept-Encoding interactions)

Why

• Better cross-provider compatibility: MinIO’s Go client targets the S3 API surface used by “S3-compatible” stores, which reduces surprises when talking to non-AWS endpoints.
• Avoids recent checksum/signing pitfalls: AWS SDK v2 introduced default response/integrity checksum behaviors that can interact poorly with some S3-compatible services, contributing to SignatureDoesNotMatch failures. Using minio-go sidesteps those defaults and focuses on portable S3 semantics.
• GCS S3 interoperability: We rely on HMAC/S3-style access against Google Cloud Storage; their interoperability path is documented, and minio-go is frequently used with such endpoints

What changes

• Implementation detail only: all object operations (HEAD/GET/PUT/COPY) are handled by minio-go
• No changes to caller-facing config/flags (bucket/prefix/path-style/endpoint)

---

Closes: #3749 (minio-go works with GCS via HMAC)
Closes: #5804 (minio-go works with OpenStack S3)
Closes: #4487 (minio-go works with Alibaba Cloud)

[tonistiigi]

cc @bpaquet