ci: drop -f from test compile flags to speed up CI

[chaitanyaprem]

Drops the -f (force recompile) flag from the shared cfg used by every runTest and buildExample invocation in libp2p_mix.nimble.

Why

Each test in nimble test / nimble testComponent is its own Nim binary, but they all share a single nimcache/ directory in the working tree. With -f present, every test recompiles its entire transitive dep tree (libp2p, chronos, stew, nimcrypto, …) from scratch — the dominant cost of CI.

Without -f, the first test pays the full compile cost; subsequent tests reuse compiled .o files from nimcache. Expected per-job reduction: ~50-70% of the test-phase wall-clock.

Baseline (from PR #14's CI run)

Stacked on top of #12 (chore/bump-libp2p-v2.0.0). Before this change, the test-phase per-job times on a fresh runner were:

Job	Unit tests	Component tests	Total
linux-amd64 (Nim 2.2.4)	13m54s	(similar)	~26-28m
linux-amd64 (Nim 2.2.10)	14m10s	(similar)	~26-28m
linux-i386 (Nim 2.2.4)	14m20s	(similar)	~26-28m
linux-i386 (Nim 2.2.10)	14m22s	(similar)	~26-28m
macos-arm64 (Nim 2.2.4)	8m56s	(similar)	~22m
macos-arm64 (Nim 2.2.10)	9m19s	12m05s	26m57s

CI for this PR should show the same number of tests run with substantially lower wall-clock — that's the validation.

Why this is safe

• All test invocations share identical compile flags. Both nimble test and nimble testComponent call runTest(name) with no per-test moreoptions. Every test compiles with the same cfg, so nimcache hits will always be flag-correct.
• The retained flags are intentional: --styleCheck:usages, --styleCheck:error, --opt:speed, --threads:on, -d:libp2p_mix_experimental_exit_is_dest. Only the recompile-everything bit is gone.
• If a future test needs different flags, pass them via moreoptions and pair with -f for that specific invocation rather than re-enabling it globally.

Diff

One file, ~9 lines added (most of which is an explanatory comment):

 let cfg =
   " --styleCheck:usages --styleCheck:error" & (if verbose: "" else: " --verbosity:0") &
-  " --skipUserCfg -f --threads:on --opt:speed" &
+  " --skipUserCfg --threads:on --opt:speed" &
   " -d:libp2p_mix_experimental_exit_is_dest"
+  # `-f` (force recompile) was previously here. Dropped so test binaries
+  # share the nimcache: each subsequent test in `nimble test` /
+  # `nimble testComponent` reuses compiled libp2p / chronos / stew object
+  # files from the first test, cutting per-test compile time substantially.
+  # Safe because every test invokes `runTest(name)` with no per-test
+  # compile flag overrides — all tests build with identical `cfg`, so
+  # cache hits are flag-correct.

Rollback

If this surfaces flaky behavior in CI (test cross-contamination via stale nimcache), the rollback is a one-character revert — re-add -f in cfg. No downstream impact since libp2p_mix.nimble's test config is purely internal to this repo.

[chaitanyaprem]

Closing — the -f removal turned out to be a no-op on the test phase.

Why it didn't work: each nim c tests/test_X.nim invocation defaults to its own nimcache/test_X_d/ subdirectory (keyed on the entry-point filename). The 14 unit + 6 component tests in a single CI job never shared compilation in the first place, so removing -f (the force-recompile flag) couldn't unlock anything.

CI on this PR vs the prior baseline (PR #14):

• linux-amd64 (2.2.10): 36m18s vs ~26-28m in-flight (no meaningful change)
• linux-i386 (2.2.10): 37m35s — same range
• macos-arm64 (2.2.10): 25m15s vs 26m57s (~6%, within runner variance)

Tracked as # for proper investigation (likely fix: explicit --nimcache:<shared-path> so all tests share compiled object files).

[chaitanyaprem]

Tracked in #17.