fix(grading): stop autograder from silently crashing under bash 5

[lassebenni]

Summary

• ((var += expr)) reports a failing exit status whenever the resulting arithmetic value is 0 -- a well-known bash gotcha. Combined with set -euo pipefail, the grader silently dies mid-run the moment a student's running score hits exactly 0 at a checkpoint (e.g. a mostly-empty early submission).
• This only shows up on GitHub Actions' Ubuntu runners (bash 5+); it's invisible when testing locally on macOS (bash 3.2 ships without this strictness), which is how it went unnoticed.
• Also guards bare check_*() helper calls (which legitimately return 1 to signal warn/fail, not a script error) with || true so they don't trip set -e either.

Test plan

• bash -n .hyf/test.sh
• Reproduced against an empty scaffold repo with GNU bash 5: exit code 1 before the fix, exit code 0 after.
• Re-ran against a known-good local solution: score/pass output unchanged (regression-free).
• Confirm a real PR run in CI now completes the "Run tests" step for a low/zero-scoring submission.

🤖 Generated with Claude Code

[github-actions]

📝 HackYourFuture auto grade

Assignment Score: 48 / 100 ❌

Status: ❌ Not passed
Minimum score to pass: 60
🧪 The auto grade is experimental and still being improved

Test Details

=== Week 10 Autograder ===
  ✓ PASS  found dbt_project.yml
  ✓ PASS  found packages.yml
  ✓ PASS  found profiles.yml.example
  ✓ PASS  found macros/safe_divide.sql
  ✓ PASS  found models/staging/_sources.yml
  ✓ PASS  found models/staging/stg_trips.sql
  ✓ PASS  found models/staging/stg_zones.sql
  ✓ PASS  found models/marts/fct_daily_borough_stats.sql
  ✓ PASS  found reports/answers.md
  ✓ PASS  found AI_ASSIST.md
  ✓ PASS  found singular test in tests/
  ✗ FAIL  docs/lineage.png missing -- run dbt docs generate + serve and screenshot the lineage graph
  ✓ PASS  Level 1: required files (0/10 pts)
  ✓ PASS  .gitignore: profiles.yml excluded
  ✓ PASS  profiles.yml.example: uses env_var() -- no hardcoded password
  ✓ PASS  profiles.yml not committed (correctly git-ignored)
  ✓ PASS  Level 2: secrets hygiene (15/15 pts)
  ✓ PASS  stg_trips.sql: uses {{ source() }} reference
  ✗ FAIL  stg_trips.sql: must filter WHERE pickup_location_id IS NOT NULL
  ✗ FAIL  stg_trips.sql: must filter WHERE fare_amount >= 0
  ✗ FAIL  stg_trips.sql: tip_pct column missing -- add {{ safe_divide('tip_amount', 'fare_amount') }} AS tip_pct
  ✓ PASS  stg_zones.sql: uses {{ source() }} and is filled
  ✓ PASS  Level 3: staging models (8/20 pts)
  ✓ PASS  fct_daily_borough_stats.sql: file filled
  ✓ PASS  mart: ref() to both stg_trips and stg_zones
  ✗ FAIL  mart: no JOIN -- must join stg_trips to stg_zones on pickup_location_id = location_id
  ✗ FAIL  mart: no GROUP BY -- must aggregate to (pickup_borough, pickup_date) grain
  ✓ PASS  mart: all 6 required output columns present
  ✓ PASS  Level 4: mart model (12/20 pts)
  ✗ FAIL  mart YAML: missing dbt_utils.unique_combination_of_columns on (pickup_borough, pickup_date)
  ✓ PASS  staging YAML: not_null tests present
  ✓ PASS  singular test: filled (does not reference fct_daily_borough_stats -- check)
  ✓ PASS  Level 5: tests (8/15 pts)
  ✗ FAIL  mart YAML: grain mentioned but description still has TODO -- fill it in
  ✗ FAIL  mart YAML: column descriptions are empty -- explain meaning and units for each column
  ✓ PASS  Level 6: documentation (0/10 pts)
  ✓ PASS  reports/answers.md: 4 SQL blocks present and filled
  ✗ FAIL  AI_ASSIST.md: empty or still contains TODO stubs
  ✓ PASS  Level 7: business answers + AI log (5/10 pts)

Score: 48 / 100  (passing: 60)  pass=false