Add [\!shouldfail] reproducer for transitive implicit-dep tracking bug

[typeless]

Replaces #48 (closed when its base branch feature/show-index was deleted on merge of #49 — --delete-branch killed the dependent PR rather than retargeting it).

Summary

Adds a minimal failing reproducer for a real bug in pup's implicit-dep recording, tagged [\!shouldfail] so CI stays green while the bug is on file. Uses pup show index (just landed via #49) for forensic diagnostic output.

The bug

When an already-tracked header is edited to #include a new header, the source's recompile produces a .d file listing the new header, but pup's index does not extend the source's implicit-dep set to include it. Subsequent edits to the newly-introduced header are then invisible to change detection — pup reports Nothing to do (up to date). while the .o is stale.

This is distinct from the scope-filter bug fixed in 315b2a2. That one was an existing recorded header being filtered out during change detection. This is a header never recorded in the first place. It fires unscoped.

Reproducer (see test/e2e/fixtures/header_dep_transitive/test.sh)

1. main.c includes only old.h. Initial build records old.h as implicit dep. .o contains 0x1 (ANSWER=1). ✓
2. Edit old.h to #include "newhdr.h" and use EXTRA. Rebuild. .o contains 0x64 (EXTRA=100). ✓ (compile happened, .d listed newhdr.h)
3. Edit newhdr.h (EXTRA 100 → 7777). Rebuild expected. Observed: Nothing to do. .o stays at 0x64. ✗

Between steps 2 and 3, test.sh runs pup show index PATTERN and reports whether newhdr.h appears in main.c's implicit-dep set. Live result on this branch:

implicit: src/main.c
implicit: /usr/include/stdc-predef.h
implicit: include/old.h          ← recorded ✓
                                 ← newhdr.h missing ✗
sticky:   src/Tupfile

This isolates the root cause precisely: pup's post-execution .d-file processing does not extend the implicit-dep set with newly-discovered transitive headers. The bug is in the dep-recording path (around discovered_deps / process_implicit_deps in cmd_build.cpp), not in change detection.

Why [\!shouldfail]

Catch2's [\!shouldfail] marks the test as an expected-failure: it runs, output is captured, exit status reports 1 failed as expected. Default CI stays green. Once the dep-recording bug is fixed and the test passes, removing the tag converts it into a regression guard.

Test plan

• Reproducer runs standalone: bash test/e2e/fixtures/header_dep_transitive/test.sh exits 1 with diagnostic output showing stale .o AND the forensic from pup show index pinpointing the missing edge
• Catch2 SCENARIO via ./build/test/unit/putup_test "[\!shouldfail]" reports 1 failed as expected
• Full test suite is green (the SCENARIO does not break CI)
• make format clean

Refs pup-header-detection-bug.md (working notes).

🤖 Generated with Claude Code