[pkg/ottl] Use generics to avoid context cast in getters and funcs

[bogdandrutu]

This PR also improves number of allocations (by 2, if I am not wrong is because of the removing of the ottl.TransformContext interface use, which requires a heap allocation) and overall performance of the transform processor.

Before:

goos: darwin
goarch: amd64
pkg: github.com/open-telemetry/opentelemetry-collector-contrib/processor/transformprocessor/internal/traces
cpu: Intel(R) Core(TM) i9-9880H CPU @ 2.30GHz
BenchmarkTwoSpans
BenchmarkTwoSpans/no_processing
BenchmarkTwoSpans/no_processing-16         	  783144	      1394 ns/op	    1528 B/op	      31 allocs/op
BenchmarkTwoSpans/set_attribute
BenchmarkTwoSpans/set_attribute-16         	  713299	      1691 ns/op	    1832 B/op	      35 allocs/op
BenchmarkTwoSpans/keep_keys_attribute
BenchmarkTwoSpans/keep_keys_attribute-16   	  789517	      1626 ns/op	    1560 B/op	      33 allocs/op
BenchmarkTwoSpans/no_match
BenchmarkTwoSpans/no_match-16              	  799210	      1520 ns/op	    1560 B/op	      33 allocs/op
BenchmarkTwoSpans/inner_field
BenchmarkTwoSpans/inner_field-16           	  737241	      1566 ns/op	    1560 B/op	      33 allocs/op
BenchmarkTwoSpans/inner_field_both_spans
BenchmarkTwoSpans/inner_field_both_spans-16         	  763725	      1622 ns/op	    1592 B/op	      35 allocs/op

After:

goos: darwin
goarch: amd64
pkg: github.com/open-telemetry/opentelemetry-collector-contrib/processor/transformprocessor/internal/traces
cpu: Intel(R) Core(TM) i9-9880H CPU @ 2.30GHz
BenchmarkTwoSpans
BenchmarkTwoSpans/no_processing
BenchmarkTwoSpans/no_processing-16         	  868864	      1296 ns/op	    1480 B/op	      29 allocs/op
BenchmarkTwoSpans/set_attribute
BenchmarkTwoSpans/set_attribute-16         	  738514	      1609 ns/op	    1784 B/op	      33 allocs/op
BenchmarkTwoSpans/keep_keys_attribute
BenchmarkTwoSpans/keep_keys_attribute-16   	  800833	      1450 ns/op	    1512 B/op	      31 allocs/op
BenchmarkTwoSpans/no_match
BenchmarkTwoSpans/no_match-16              	  835160	      1394 ns/op	    1512 B/op	      31 allocs/op
BenchmarkTwoSpans/inner_field
BenchmarkTwoSpans/inner_field-16           	  796388	      1445 ns/op	    1512 B/op	      31 allocs/op
BenchmarkTwoSpans/inner_field_both_spans
BenchmarkTwoSpans/inner_field_both_spans-16         	  798840	      1539 ns/op	    1544 B/op	      33 allocs/op

[TylerHelmuth]

I'll do an in-depth review over the next week, but no feedback so far.

[TylerHelmuth]

Maybe a followup change, but instead of a generic NewParser, should contexts provide a Parser specific to their context? Something like ottllogs.NewParser(logs.Functions(), component.TelemetrySettings{})?

[bogdandrutu]

I like that more, since 2 arguments are always coming from that package. Will address it in the next PR.

[TylerHelmuth]

Sweet, do you plan to work on that or do we need an issue?

[bogdandrutu]

Will do after you approve this 🗡️

[TylerHelmuth]

Maybe we could do something similar here as proposed for NewParser? Have []ottllogs.Statements so that we don't have to have the processor pass in the Generic itself.

Hopefully we can make it so that the only place a processor would need to directly interact with the generic is when creating its own functions.

[bogdandrutu]

So essentially have a TracesStatements?

[bogdandrutu]

I can also look into this for the next PR when I look into the Parser.

[TylerHelmuth]

Ya something like that. Since the generic introduce some complexity, will be great to hide that complexity inside the Contexts as much as possible since, at least at the moment, we need a Parser and set of Statements per context.

[TylerHelmuth]

@bogdandrutu I'll give this a final look over tomorrow.

[evan-bradley]

Looks good to me. Nice use of generics!