BatchSpend:

• Do not attempt to SET 0 new TATs, this is an edge-case which only impacted testing and triggers a divide-by-zero panic in "ratelimits_latency" metric calculation.

Metric "ratelimits_latency":

• Correct an issue where a single non-existent key in a "batchget" operation incorrectly emits both a status="notFound" and an additional status="success" metric for the same operation.
• Emit status="mixedSuccess" status for "batchget" operations where some entries are not found
• Emit a "batchget_entry" for each entry in a "batchget"
• Emit a "batchset_entry" for each entry in a "batchset"
• Calculate latency using r.clk.Since() instead of time.Since(), this should only matter for tests.

Metric "ratelimits_spend_latency":

• Include BatchGet latency in the overall "ratelimits_spend_latency", this was omitted accidentally.
• Emit latency and outcome per transaction. We're going to need this to compare apples-to-oranges with each of the legacy rate limits.

Fixes #7597