oximeter: optimize measurements tables.

[jmcarp]

Apply a few ddl-only optimizations to the oximeter measurements tables:

• Encode timestamps columns using delta encoding. Rows are sorted by (timeseries_name, timeseries_key, timestamp), so timestamps compress very well as deltas.
• Partition measurements tables by timestamp, within days. Many queries care about a small time range, and partitioning by day will make those queries significantly cheaper.

Note: this is a draft, since I need to better understand how clickhouse migrations work, and how to do this safely. I'll apply these changes to all measurements tables if/when we agree to an approach and convince ourselves that this will be safe to run.

[jmcarp]

I dug into this a bit more, pulling a few known slow queries (to make sure they got faster) and a known fast query (to make sure it didn't get slower). On closer inspection, partitioning seems pretty useful, speeding up the slower queries by over 50%, but changing the timestamp encoding actually makes us go slower. I'll update this PR to add partitioning and leave encoding alone.

• virtual_machine:vcpu_usage: ~0.515s | ~0.18s | ~0.5s
• hardware_component:current: ~1.8s | ~0.52 | ~0.66s
• switch_data_link:fec_high_symbol_errors : ~0.03s | 0.02s

Note: I talked with @bnaecker about this, and if we decide to go ahead with the partitioning change, we'll need to wait on #7488. Until that issue is resolved, we effectively have no good way to update clickhouse schemas.