fix(install): regenerate lockfile with --force on global install - #33970

Fix is well-scoped — deletion is limited to dir.join("deno.lock") inside the per-install config dir (~/.deno/bin/.<name>/), and the if lockfile_path.exists() guard handles the fresh-install case cleanly.

Two test nits:

• install_reload_regenerates_lockfile sets reload: true and force: true. Since force || flags.reload short-circuits on the first true, the test doesn't actually exercise the reload-only path — passing force: false here would isolate it.
• The same test calls rustls::crypto::aws_lc_rs::default_provider().install_default() while install_force_regenerates_lockfile doesn't. Both fetch from http://localhost:4545 (no TLS), so the rustls init looks cargo-culted — unless there's a code path I'm missing that lands on a TLS request.

Holding approval until CI is green (12 shards still in_progress at the head SHA).

fdcb6e9d narrows the condition from force || flags.reload → force only. The linked issue #33934 explicitly lists both flags in the expected behaviour ("When --force (or --reload) is specified... the lockfile should be regenerated"). Was the reload removal intentional, or a misread of my prior nit about test isolation? The nit was "make the reload test actually exercise reload-only", not "drop reload from scope."

If --force-only is what you want, two test follow-ups:

• install_reload_regenerates_lockfile still has force: true, so it's now a misleadingly-named force test. Either rename it or convert it to assert that force: false, reload: true does NOT remove the lockfile — that pins the new (narrowed) behaviour and prevents accidental reflow back to force || reload.
• The rustls::default_provider().install_default() call in that test is now even more orphaned (no TLS path at all).

Holding approval until the scope question is settled and CI lands green.

537d8416 renames the test to install_reload_preserves_lockfile, flips force: false, asserts post_reload_content == stale_lockfile, and drops the orphaned rustls init. That pins the narrowed --force-only scope and prevents accidental drift back to force || reload — both prior nits resolved and the scope question is answered by the test itself.

Heads-up on the linked issue: #33934's expected-behaviour section still names both flags, so if @bartlomieju wants --reload to also clear the lockfile that's a maintainer call — code is internally consistent either way.

Holding approval until CI lands green on 537d8416 (most shards still in_progress).