chore: fixing problem with load event and performance for loadend

[obecny]

Which problem is this PR solving?

It appeared that load event might not yet have the performance metrics for loadEnd The support for loadEnd event is very limited and cannot be used instead. The solution is to use timeout when document loads so that this one missing performance metric is in place.

[codecov-io]

Codecov Report

Merging #469 into master will increase coverage by 0.17%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #469      +/-   ##
==========================================
+ Coverage   95.19%   95.37%   +0.17%     
==========================================
  Files         137      137              
  Lines        6833     7003     +170     
  Branches      594      598       +4     
==========================================
+ Hits         6505     6679     +174     
+ Misses        328      324       -4

Impacted Files	Coverage Δ
...telemetry-plugin-document-load/src/documentLoad.ts	97.75% <100%> (+0.07%)	⬆️
...entelemetry-plugin-postgres/test/assertionUtils.ts	96.55% <0%> (-3.45%)	⬇️
...opentelemetry-base/test/resources/resource.test.ts	100% <0%> (ø)	⬆️
...entelemetry-exporter-jaeger/test/transform.test.ts	100% <0%> (ø)	⬆️
...ges/opentelemetry-exporter-zipkin/src/transform.ts	100% <0%> (ø)	⬆️
.../opentelemetry-exporter-zipkin/test/zipkin.test.ts	100% <0%> (ø)	⬆️
.../opentelemetry-exporter-jaeger/test/jaeger.test.ts	100% <0%> (ø)	⬆️
...telemetry-scope-base/test/NoopScopeManager.test.ts	100% <0%> (ø)	⬆️
packages/opentelemetry-web/src/WebTracer.ts	100% <0%> (ø)	⬆️
...entelemetry-exporter-zipkin/test/transform.test.ts	100% <0%> (ø)	⬆️
... and 3 more

[draffensperger]

Would it make sense to have a little delay here, e.g. 1 second or so?

[obecny]

if I understand correctly it should work like this. When load event is being called then the performance is collecting the remaining metrics included "loadEnd", I assume this is synchronous so whenever this is finished the browser will start executing all the timeouts that were set together with the same load event. Then it really doesn't make sense to wait longer as the loadEnd metric should be there. If this is not the case then it will be hard to really say how long you should wait, 100ms, 1s, maybe 2s, but it seems like the moment the setTimeout calls the callback this metric is already there.

[obecny]

In the end this is not final, we can always put there 1s at any moment if for any reason it appeared that some browsers don't have it yet.