Split task run metrics into various dimensions to allow more narrow SLOs

[mikecote]

Problem Statements

• Systematic errors are all part of the same bucket, preventing us from creating dedicated SLOs at various service levels (task manager, alerting framework and actions framework)
• Rule type implementers cannot create SLOs on their rule types because systematic errors may be related to the framework and not their code
• We set low SLO targets to accomodate user errors, while Task Manager service itself could benefit from a higher target (ex: 99.999)

Proposed Solution

TBD @mikecote

[elasticmachine]

Pinging @elastic/response-ops (Team:ResponseOps)