Change expensive timer metrics to trend counters

[SirTyson]

Description

Resolves https://github.com/stellar/stellar-core-internal/issues/336

Changes expensive timer metrics to counters.

For timer and histogram metrics, we store data points in a buffer. When this buffer "fills up" we must to a compaction step, which involves many copies. For metrics that measure short lived functions (runtime in the ns or us range) that get called very frequently, the cost of this compaction step dominates the function runtime. For example, for BucketList in memory loads, metric overhead accounted for 75% of the load's total runtime.

This changes several short lived metrics to trends instead of timer based histograms. Specifically, we no longer track individual data samples, but keep a running sum of runtime instead. This still gives us insight into trends, but no longer gives us info on individual calls. However, it is significantly cheaper runtime wise.

Checklist

• Reviewed the contributing document
• Rebased on top of master (no merge commits)
• Ran clang-format v8.0.0 (via make format or the Visual Studio extension)
• Compiles
• Ran all tests
• If change impacts performance, include supporting evidence per the performance document

[marta-lokhova]

LGTM, just one small typo. Feel free to squash the last commit and rebase.

[SirTyson]

LGTM, just one small typo. Feel free to squash the last commit and rebase.

Done