Benchmark: optimize by parallel builds and caching

[liorsve]

Added optimizations for the benchmark workflow.

benchmark.yml changes

• Parallel background builds
  Replaced 5 sequential steps (Build SDV, Install glide deps, Build glide, Build resp-bench, Install deps) with a single step that runs all builds concurrently as background processes:

  • Valkey server compilation
  • SDV build from source (if needed and not cached)
  • Glide build from source (if needed and not cached)
  • pip install
  • async-profiler download

• Build caching (keyed by commit ID / version, persists across runs)

  • SDV build artifacts (sdv-build-{commit})
  • Glide build artifacts (glide-build-{commit})
  • Valkey server binary (valkey-server-{version})
  • Maven local repository (maven-bench-{resp-bench-commit})
  • async-profiler (async-profiler-3.0)

• Consolidated apt-get
  Merged 3 separate apt-get update + install calls into a single step that installs all system packages upfront
  (Java, Maven, pip, sysstat, numactl, gcc, cmake, etc.)

• Unified Glide build
  Merged duplicate "Build valkey-glide from source (primary/secondary)" steps into a single step with a unified build_glide flag that handles both cases.

Timing Results

Tested on opt-bench-1-runner-clean branch.

#	Scenario	Overall Parallel Builds step	Parallel Build step details	resp-bench build	Benchmark	Job exec	Queue	Total (Queue+Job exec)
1	SDV + Glide defaults (no builds)	0m04s	cached	0m12s	1m08s	1m58s	1m53s	3m51s
2	SDV commit (cache miss)	1m54s	sdv 1m54s	0m11s	0m39s	3m18s	1m53s	5m11s
3	Glide commit (cache miss)	7m47s	glide 7m47s	0m12s	0m42s	9m14s	1m53s	11m07s
4	SDV + Glide commits (cache misses)	7m48s	sdv ~2m, glide 7m48s	0m12s	0m41s	9m17s	1m53s	11m10s
5	Jedis version override	0m04s	cached	0m12s	0m32s	1m22s	1m53s	3m15s
6	SDV + Glide release versions (no builds)	0m04s	cached	0m12s	1m07s	2m03s	1m53s	3m56s
7	SDV commit (cached)	0m04s	all cached	0m12s	0m39s	1m26s	1m53s	3m19s
8	SDV + Glide commits (cached)	0m05s	all cached	0m12s	0m40s	1m27s	1m53s	3m20s

[jeremyprime]

Note that GLIDE requires JDK 11 (otherwise it will fallback to using Foojay, which is down a lot, causing the benchmark run in the CI to fail). We may want to install JDK 21 and JDK 11 in this step, then in GLIDE-specific steps set JAVA_HOME to use JDK 11. See #75 for an example.