Return Span from DefaultSpan factories.

[anuraaga]

#1135 talks about hiding DefaultSpan completely. I couldn't quite find a good home for the factory methods (maybe DefaultTracer?) but it seems like a good idea.

In the meantime, it's more important to make the return type of the factories generic because users should definitely not have to care whether a span is a DefaultSpan or not. In practice, the return value of DefaultSpan makes it difficult to bridge them in auto instrumentation.

open-telemetry/opentelemetry-java-instrumentation#764

[codecov]

Codecov Report

Merging #1519 into master will not change coverage.
The diff coverage is n/a.

@@            Coverage Diff            @@
##             master    #1519   +/-   ##
=========================================
  Coverage     87.01%   87.01%           
  Complexity     1367     1367           
=========================================
  Files           162      162           
  Lines          5183     5183           
  Branches        490      490           
=========================================
  Hits           4510     4510           
  Misses          492      492           
  Partials        181      181           

Impacted Files	Coverage Δ	Complexity Δ
.../main/java/io/opentelemetry/trace/DefaultSpan.java	80.00% <ø> (ø)	18.00 <0.00> (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 bcb1d4a...5529657. Read the comment docs.

[carlosalberto]

LGTM. I've long wondered whether there's any reason to expose DefaultSpan publicly. @bogdandrutu may know better (although I can't think of one at the moment).