fix(benchmark): remove personal SWE-bench path defaults

[beruro]

Problem

src-tauri/src/benchmark/mod.rs hardcoded a developer's personal absolute path (/Users/laptop-h/Documents/GitHub/SWE-bench_Pro-os) as the production default for the SWE-bench Pro harness repository (evaluator script + per-task run scripts). On any other machine this default is guaranteed wrong, it leaks a personal directory layout into the codebase, and a missing harness only surfaced late as a file-not-found style failure instead of a clear "configure first" state.

Solution

• Deleted the personal-path default constant entirely; there is no filesystem default anymore.
• Extracted resolution into a pure function resolve_swe_bench_repo_path(explicit, env_value) in benchmark/paths.rs with explicit precedence: explicit value (app setting / command argument, reserved — no caller passes one yet) → ORGII_SWE_BENCH_PRO_REPO_PATH environment variable → typed SweBenchRepoNotConfigured error. Blank/whitespace values count as unset; resolved values are trimmed.
• The existing ORGII_SWE_BENCH_PRO_REPO_PATH semantics are preserved (the E2E suite fixture in tests/e2e/wdio.conf.mjs keeps working unchanged).
• swe_bench_evaluator_script_path() / swe_bench_run_scripts_dir() now return Result, so:
  • benchmark_preflight reports the unconfigured state as failing checks (evaluator_script, run_scripts_dir) whose detail is an actionable message naming the environment variable — the frontend preflight panel shows "configure first" up front;
  • benchmark_create_run_plan / benchmark_start_run in local_docker mode fail immediately at plan-build time with the same clear message instead of spawning an evaluator that dies on a nonexistent path.
• swe_bench_command_preview now receives the already-resolved paths instead of re-resolving them internally.
• Unit tests in benchmark/paths.rs cover the precedence contract: explicit wins over env, env fallback, blank-counts-as-unset, trimming, typed not-configured error, and that the error message names the env var.

Potential risks

• Machines that relied on the old personal-path default silently "working" (only the original developer's machine) must now set ORGII_SWE_BENCH_PRO_REPO_PATH. This is intended behavior, not a regression.
• There is no in-app settings UI for the harness path yet; the env var is the only configuration source. Follow-up: add an app-settings field and pass it through the reserved explicit parameter of resolve_swe_bench_repo_path.
• Behavior change: in local_docker mode, run-plan creation now errors out when unconfigured instead of returning a plan with a failing preflight. Frontend surfaces the returned error string; no DTO shape changed.
• Local compile verification did not complete on this machine (see Verification); CI must gate this PR.

Verification

• rg -n "/Users/laptop-h|/Users/junyu" src-tauri/src/benchmark/ → empty (exit 1, no matches).
• cargo fmt -p org2 --check → the only diffs reported are pre-existing unformatted files elsewhere in the crate (untouched by this PR); the four changed benchmark files are clean.
• cargo check -p org2 --lib → NOT RUN to completion: full cold compile of the app crate repeatedly exceeded the available time budget on a machine already running several parallel builds. Relying on CI for compile + test gating — hence Draft.
• cargo test -p org2 --lib (new benchmark::paths::tests) → NOT RUN for the same reason; the tests are pure-function assertions with no I/O.