fix(valgrind): tie --separate-threads to subprocess tracking

[not-matthias]

--separate-threads was keyed on --cycle-estimation, which is unrelated: cycle estimation changes how event costs are computed, not how they are partitioned across threads. As a side effect, turning cycle estimation off silently also merged thread data, and turning it on split it.

Per-thread dumps only matter when instrumentation is inherited across a traced exec (--instr-atstart=inherit), so the option now follows simulation_track_subprocess and is set explicitly in both branches:

• tracking subprocesses: --instr-atstart=inherit --separate-threads=yes
• otherwise: --instr-atstart=no --separate-threads=no (valgrind's default, so no behavior change for existing runs)

The flag now sits in the tool-agnostic branch rather than the callgrind arm. That is safe: tracegrind is a fork of callgrind and accepts --separate-threads in its own option parser, same as callgrind.

[codspeed-hq]

Merging this PR will not alter performance

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

✅ 17 untouched benchmarks

---

_{Comparing valgrind-separate-threads-subprocess (1405d07) with main (3798897)}