feat: prometheus serializer

[legendecas]

Which problem is this PR solving?

• Fixes Remove prom-client from prometheus exporter #1039

Short description of the changes

• Add PrometheusSerializer to serialize aggregated metric records.
• Add PrometheusLabelsBatcher to aggregate metric records with different labels.
• Remove dependency on prom-client.

[codecov]

Codecov Report

Merging #1310 into master will increase coverage by 0.10%.
The diff coverage is 94.47%.

@@            Coverage Diff             @@
##           master    #1310      +/-   ##
==========================================
+ Coverage   93.80%   93.91%   +0.10%     
==========================================
  Files         149      151       +2     
  Lines        4539     4615      +76     
  Branches      940      953      +13     
==========================================
+ Hits         4258     4334      +76     
  Misses        281      281              

Impacted Files	Coverage Δ
...etry-exporter-prometheus/src/PrometheusExporter.ts	89.83% <89.83%> (ø)
...ry-exporter-prometheus/src/PrometheusSerializer.ts	96.15% <96.15%> (ø)
...exporter-prometheus/src/PrometheusLabelsBatcher.ts	100.00% <100.00%> (ø)

[vmarchaud]

overal lgtm, nit pick about comparison

[obecny]

1. I think you are changing here too many things at the same time.
2. The serializer itself is not spec compliant
3. The changes to metrics should be done first in separate PR, after that PR you can focus on serialisation only .

[dyladan]

@obecny have your concerns here been addressed?

[obecny]

last concern with the batcher - added comment

[legendecas]

#1428 added mapping MetricKind. VALUE_RECORDER to Prometheus data type as Gauge. However, the aggregator type by default determined by Batcher is MinMaxLastSumCountAggregator for MetricKind. VALUE_RECORDER, which in otel collector exporter is been mapped as Summary. To keep alignment, aggregator MinMaxLastSumCountAggregator is mapped as summary in the latest changes.

[obecny]

#1428 added mapping MetricKind. VALUE_RECORDER to Prometheus data type as Gauge. However, the aggregator type by default determined by Batcher is MinMaxLastSumCountAggregator for MetricKind. VALUE_RECORDER, which in otel collector exporter is been mapped as Summary. To keep alignment, aggregator MinMaxLastSumCountAggregator is mapped as summary in the latest changes.

Choosing the aggregation based on metric type might be wrong. The user can use the same metric and create its own batcher and use different aggregators for the same metrics. Also due to last conversation on spec we will have to change ValueObserver aggregator and ValueRecorder again so to avoid any such problems in future we should only depends on aggregator type and never on metric type when trying to get the value (btw MinMaxLastSumCountAggregator will be removed completely)

[obecny]

lgtm

[dyladan]

This looks ok to me, but I want to mention that this looks like it does a lot more than just the serialization. In the future, I would appreciate if you would break out other work into separate PRs because it makes it very difficult to review when the PR is so large.

I am also concerned that many of the largest changes seem to have come after the first round of reviews, which means the approvals it already had were for code that is very different than the current state.

I am not going to block this PR, but please keep these things in mind for future PRs.

[legendecas]

Thanks for the reviews! :D