[PROF-11405] Graduate heap profiling from alpha to preview, second try

[ivoanjo]

What does this PR do?

A few days ago, we merged #4331 and #4401 to master.

Starting with #4331, we used a new libdatadog feature for implementing the Ruby heap profiler. While validating these changes, we discovered this libdatadog feature had a bug (DataDog/libdatadog#896) and thus reverted all PRs related to this heap profiler change until we could get the fix in #4409.

Because the bug was only in the new libdatadog feature, it was still OK to use the old approach with the buggy libdatadog (since it used an alternative, non-buggy, code path).

Now that we have the fix on the libdatadog side almost ready, I'm opening this PR to re-apply the changes from #4331 and #4401 to master.

This PR is thus almost the reverse of the revert PR in #4409. The only difference is that it does not yet include the changes from #4376 for the same reason that PR was needed originally.

Motivation:

Release heap profiling feature.

Change log entry
Yes. Please use the one in #4401 .

Additional Notes:

There's quite a diff on this PR, but there's no differences to those earlier PRs other than adding a regression test for this issue (the regression test is a separate commit that can be reviewed by itself).

In fact, you may notice that CI is red for this PR because I'm opening it before the libdatadog fix is released. My intention is to exactly come back once the new libdatadog is released, and see this PR turn green with a simple merge from master.

How to test the change?

The earlier PRs already included tests; and I added a regression test for the libdatadog issue in this PR.

[datadog-datadog-prod-us1]

Datadog Report

Branch report: ivoanjo/prof-11394-heap-profiling-graduate-try2
Commit report: f850521
Test service: dd-trace-rb

❌ 1 Failed (0 Known Flaky), 19818 Passed, 1366 Skipped, 3m 16.53s Total Time

❌ Failed Tests (1)

• Datadog::Profiling::StackRecorder libdatadog managed string storage regression test when reusing managed string ids across multiple profiles produces correct profiles - rspec - Details

  Expand for error

   expected ["", "local root span id", "trace endpoint", "end_timestamp_ns", "cpu-time", "nanoseconds", "cpu-samp...lloc-samples", "alloc-samples-unscaled", "heap-live-samples", "heap-live-size", "bytes", "timeline"] to include "key", "hello", and "world"
   
   Failure/Error: expect(decoded_profile2.string_table).to include("key", "hello", "world")
     expected ["", "local root span id", "trace endpoint", "end_timestamp_ns", "cpu-time", "nanoseconds", "cpu-samp...lloc-samples", "alloc-samples-unscaled", "heap-live-samples", "heap-live-size", "bytes", "timeline"] to include "key", "hello", and "world"
   ./spec/datadog/profiling/stack_recorder_spec.rb:1095:in 'block (4 levels) in <top (required)>'
   ./spec/spec_helper.rb:238:in 'block (2 levels) in <top (required)>'
   ./spec/spec_helper.rb:123:in 'block (2 levels) in <top (required)>'
   /usr/local/bundle/gems/webmock-3.25.0/lib/webmock/rspec.rb:39:in 'block (2 levels) in <top (required)>'
   /usr/local/bundle/gems/rspec-wait-0.0.10/lib/rspec/wait.rb:47:in 'block (2 levels) in <top (required)>'
  

[Strech]

LGTM 👍🏼