internal: Improve the Iter performance of LockMap on empty maps

[ajgajg1134]

What does this PR do?

Improves LockMap.Iter performance for empty maps

goos: darwin
goarch: arm64
pkg: gopkg.in/DataDog/dd-trace-go.v1/internal
                 │ original.out │          withAtomic2.out           │
                 │    sec/op    │   sec/op     vs base               │
Iter/original-10   14.280n ± 1%   1.998n ± 0%  -86.00% (p=0.002 n=6)

                 │ original.out │        withAtomic2.out        │
                 │     B/op     │    B/op     vs base           │
Iter/original-10     0.000 ± 0%   0.000 ± 0%  ~ (p=1.000 n=6) ¹
¹ all samples are equal

                 │ original.out │        withAtomic2.out        │
                 │  allocs/op   │ allocs/op   vs base           │
Iter/original-10     0.000 ± 0%   0.000 ± 0%  ~ (p=1.000 n=6) ¹
¹ all samples are equal

Motivation

Turns out doing RLock and RUnlock can still be slow especially when this code is in a HOT path. For most users this map will be empty. Therefore it's worth speeding up that case at the cost of a tiny bit of extra memory and a bit of complexity.

Describe how to test/QA your changes

Added a thrash test that when run with the race detector should help make sure we don't miss anything (and it verifies that the count and len match at the end to make sure they don't get out of sync).

Reviewer's Checklist

• Changed code has unit tests for its functionality.
• If this interacts with the agent in a new way, a system test has been added.

[pr-commenter]

Benchmarks

Benchmark execution time: 2023-07-19 15:59:01

Comparing candidate commit 5b7943f in PR branch andrew.glaude/FasterLockMap with baseline commit f905f70 in branch release-v1.53.x.

Found 0 performance improvements and 0 performance regressions! Performance is the same for 27 metrics, 0 unstable metrics.