Support single bucket for experiments

[mcopik]

We shouldn't use more than three buckets; for code, experiments, and benchmark data.

• [x] Use a dedicated name for resources, retained in cache across invocations.
• [x] Recover the resource name from existing deployments.
• [x] Support a single bucket for experiments.
• [x] Support a single bucket for all benchmark data.
• [x] Support a single bucket for code (411 on AWS).

[mcopik]

We need different approaches depending on the platform.

On AWS, we list buckets and identify deployment identifier from the bucket name. On Azure, we list resource groups and identify deployment ID there. Then, we create storage accounts and containers like always.