Delete the ghost-paths fallback the record reader now makes loud

[typeless]

merge_out_of_scope_commands scanned the new index for entries whose
non-empty path was missing from path_to_id, and its path lookup fell back
to that scan. Every producer of such an entry registers it: the two
serialization arms that skip registration are the ones that write an empty
path deliberately, keeping a slot so ids stay dense, and the loop's own
guard excluded exactly those. The two predicates cannot drift because
interning the empty string yields the empty handle the guard tests for --
which is now stated beside is_empty and pinned, since the guards rest on it
permanently while the comment that mentioned it goes with the map.

The map is not merely unexercised in practice but dominated in principle.
It can only matter on a run where a lookup consults path_to_id for an
unregistered non-empty path, and on exactly those runs deleting it is
better: resolve_file takes its miss branch instead, mints a second entry
for that path copying the old entry's type, and prior_paths has refused a
record naming one path twice since #382 -- usually through its per-list
repeats scan rather than the cross-class overlaps one, because the mint
copies the type of the entry it duplicates. So the deletion never reduces
detection and strictly increases it: what the fallback absorbed in silence
now reaches a reader.

The qualification worth stating: that loudness arrives one build late, and
as a record reported unreadable rather than as the producer that forgot to
register. It is a better failure than none, not a diagnosis.

Measured rather than argued: the function runs 234 times across the test
suite, walking 3126 file entries, and the map receives an entry zero times.
It is entered on every build that has a prior record -- the condition is
if (old_index), with no scope test -- so despite its name it is not
scoped-build-only machinery, and the incremental scenarios exercise it far
more broadly than the issue's suggested net implies.

The pin is discharged by proof by drift: deleting intern()'s empty-string
special case makes it fail. The pair review found no blockers.

Closes #391

[github-actions]

PR metrics

Performance (gcc example, Linux)

Workload	Instructions	CPU time	Page faults	D1 miss	LL miss	Wall	Peak RSS
parse	1762 M	0.64 s	14.5 k	0.6% (-0.1pp)	0%	0.649 s	32.3 MB (+0.1MB)
dry-run	2255 M	0.72 s	15.5 k	0.8%	0%	0.727 s	36.1 MB

Deterministic signals: instructions (cachegrind-simulated instruction reads — exact across runs, no PMU needed), page faults, peak RSS, and the cachegrind D1/LL miss rates. CPU time is user+sys from time(1).

Internal statistics (gcc example, up-to-date dry run)

Metric	Value
Tupfiles parsed	24
Commands	3545
Commands scheduled	0
Files checked	5834
Files changed	0
Files in index	6102
Graph edges	381415
Index size (bytes)	7683630
Implicit deps	344126
Hash computations	197
Hashes skipped (stat cache)	5636
Stat calls	5885
Parse time (ms)	573.5
Total time (ms)	714.6
Runner CPU	AMD EPYC 7763 64-Core Processor

Counters from putup -n --stat on the fully-built gcc example (up-to-date dry run): deterministic work measures — a jump in commands scheduled, hash computations, or stat calls is a real behavior change, not noise. Timings are the minimum over repeated runs, compared only against a baseline from the same CPU model; the counters are the regression signal.
Timing deltas suppressed: baseline ran on different hardware (AMD EPYC 9V74 80-Core Processor).

Binary size (Linux)

Binary	.text	.data	.bss	File
putup	562 KB (-0.1%)	2.3 KB	98.7 KB	666.4 KB (-0.6%)

Code churn (whole codebase, last 30d)

Files	Lines written	Still present	Churned	Churn rate
144	11185	9579	1606	14.4%

Of the lines written across the codebase in the last 30 days, how many are already gone — work that was written and then discarded or rewritten inside the same window. This is the state of the tree including this PR, not a measure of the PR itself. Only code we write is counted: tests, examples, vendored and generated files, CI plumbing and prose are excluded. 5206 lines were deleted in the window in total, most of them older than it.

Where the churn is

File	Lines written then discarded
src/cli/cmd_build.cpp	653
src/graph/scanners/clang_cl.cpp	110
src/cli/context.cpp	80
src/graph/builder.cpp	72
src/core/layout.cpp	69
src/core/new_delete.cpp	63
src/graph/scanners/dep_words.cpp	61
src/platform/file_io-posix.cpp	42
include/pup/index/format.hpp	35
src/cli/cmd_clean.cpp	35

Test coverage (lines)

Overall	Median file	Min file	Max file
88%	96.45%	14.7% include/pup/parser/token.hpp	100.0% include/pup/core/arena.hpp

102 files · 16430/18666 lines covered

Deltas vs main@e65214da5.

Updated for 8f20d29