Move collecting and flushing telemetry to after final request flush

[bwoebi]

Moving things around, basically between request flush and request memory free.

Reduces the request latency, even though it doesn't change anything about CPU time used.

Note: This does not affect the cli-server sapi.

[codecov-commenter]

Codecov Report

Merging #2555 (bc54d0e) into master (2e1ed12) will decrease coverage by 0.04%.
The diff coverage is 70.96%.

Additional details and impacted files

@@             Coverage Diff              @@
##             master    #2555      +/-   ##
============================================
- Coverage     75.91%   75.88%   -0.04%     
  Complexity     2574     2574              
============================================
  Files           240      240              
  Lines         27029    27069      +40     
  Branches        976      976              
============================================
+ Hits          20519    20541      +22     
- Misses         5990     6008      +18     
  Partials        520      520              

Flag	Coverage Δ
appsec-extension	69.13% <ø> (ø)
tracer-extension	78.62% <70.96%> (-0.08%)	⬇️
tracer-php	75.08% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

Files	Coverage Δ
ext/ddtrace.h	62.50% <ø> (ø)
ext/telemetry.c	96.92% <89.47%> (-3.08%)	⬇️
ext/ddtrace.c	72.61% <62.79%> (-0.53%)	⬇️

---

Continue to review full report in Codecov by Sentry.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 2e1ed12...bc54d0e. Read the comment docs.

[pr-commenter]

Benchmarks

Benchmark execution time: 2024-03-06 12:34:48

Comparing candidate commit bc54d0e in PR branch bob/remove-telemetry-from-request with baseline commit 2e1ed12 in branch master.

Found 0 performance improvements and 2 performance regressions! Performance is the same for 180 metrics, 0 unstable metrics.

scenario:SamplingRuleMatchingBench/benchRegexMatching1

• 🟥 execution_time [+187.168ns; +375.032ns] or [+3.064%; +6.139%]

scenario:SamplingRuleMatchingBench/benchRegexMatching3-opcache

• 🟥 execution_time [+143.538ns; +284.462ns] or [+2.387%; +4.731%]

[bwoebi]

This approach is problematic because of php/php-src#10335 (comment). Maybe we can use it once it's somehow fixed upstream?