Keep the stores open all the time

[badboy]

A "store" is a handle to the database.
Glean has 3 different ones and can keep them open throughout its lifetime.

Not frequently re-opening database handles also might be a performance
improvement.

Extracted from #664 because it's unrelated. Not directly related to a filed bug.

[codecov-io]

Codecov Report

Merging #665 into master will increase coverage by 0.03%.
The diff coverage is 55.26%.

@@             Coverage Diff              @@
##             master     #665      +/-   ##
============================================
+ Coverage     74.65%   74.69%   +0.03%     
  Complexity      266      266              
============================================
  Files            96       96              
  Lines          5926     5931       +5     
  Branches        730      725       -5     
============================================
+ Hits           4424     4430       +6     
- Misses          966      969       +3     
+ Partials        536      532       -4

Impacted Files	Coverage Δ	Complexity Δ
glean-core/src/database/mod.rs	62.79% <55.26%> (+0.49%)	0 <0> (ø)	⬇️
...etry/glean/private/CustomDistributionMetricType.kt	97.67% <0%> (+0.05%)	12% <0%> (ø)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 53fb434...6c8e416. Read the comment docs.

[travis79]

Out of curiosity, what was the performance gain in doing this (or do we know yet)?

[badboy]

We didn't know at submission time, it was mostly a hunch from Dexter.
I did ran a very simplistic benchmark (setting 2 metrics and submitting a ping) and get reliable 5% improvements compared to what's in current master.

glean set and submit    time:   [94.437 us 94.502 us 94.577 us]
                        change: [-5.3865% -5.1006% -4.8247%] (p = 0.00 < 0.05)
                        Performance has improved.

That is on my Linux machine (on my Mac it's unreliable and significantly slower, milliseconds instead of microseconds)

I'll work on adding the benchmark as part of this as well.

[badboy]

So the numbers are unreliable even on my Linux workstation. The benchmark is also very simplistic and nowhere near a real use case.
Nonetheless I included the benchmark now as an extra crate so we can re-run/extend it if needed.

[badboy]

holding back on this to investigate what this means for I/O operations.

[badboy]

All good, I/O operations unchanged for the startup case.

/cc @travis79, please give it a final review

[travis79]

Just a minor nit but looks good to me!