Suppress the unscanned-compile report only for a compile that scans

[typeless]

What was wrong

The unscanned-compile diagnostic exists to say when a compile's headers go unrecorded. Its
suppression — "this command writes its own depfile, so it needs no scan" — asked whether a depfile
flag appeared anywhere in the rule's command text, and was consulted per rule, before any scan
was built.

So the diagnostic was defeatable by placement rather than meaning. Any word spelling a depfile
flag silenced the report for every object the rule declared, and the failure was double-silent: no
scan, no depfile, and no report that either was missing.

The part the issue did not anticipate

Re-spelling the predicate over the scannable prefix closes the trailing and empty-prefix vectors,
but leaves the bug alive in miniature. scannable_prefix deliberately continues past
scan-transparent invocations (echo, true, :) so the walk can reach a later compile;
build_dep_scans then skips exactly those, for want of a compiler index. Reading the prefix alone
therefore still counts a flag that no scan could ever have been built from:

: a.c |> echo -MD && gcc -c a.c -o a.o && cp a.o b.o |> a.o b.o

answered "carries a depfile flag" — a program that cannot write a depfile suppressing a compile that
carries no flag. b.o is left uncovered and unreported.

Both predicates now gate on the same condition build_dep_scans gates on: the invocation must be
the compile itself. The two read the identical set of invocations, so they cannot disagree.

Also in this change

The word test moves from the command text to the tokenized words. Matching -M inside the text
required a manual check that the preceding character was a space; over words that is the
tokenizer's job, and dropping it is what retires the string_utils include here. clang_cl gets
the same shape, so the two scanners state the rule the same way rather than in two spellings.

Spec

REQ-SCAN-REPORT-UNSCANNED now reads "no invocation a scan would have been built from carries a
depfile flag" in place of "carries no depfile flag anywhere in its text", and the section's "two
classes escape the report" is down to one — the compile-and-link case, which is unrelated and
still open.

Three discharge tests, pinning both directions: a flag outside the scannable prefix suppresses
nothing, a flag inside it still suppresses, and an E2E scenario for the object left uncovered.

Verification

RED quoted for every leg, including the three second-round tests that failed against the first fix.
The E2E leg was authored after that fix, so it was proved discriminating by restoring the
pre-tightening scanner, observing the failure, and putting it back — verified by sha256, with no
git checkout and no stash.

Gate	Result
build	0
format	0
tidy	0
iwyu	0
spec-check	0
make test	0 carried onto the #397 base (838 cases, 32 shards)

On a main base make test reports the pre-existing test_builder.cpp failure from a read-only
/tmp on the development machine — the #397 defect, fixed by PR #406, in a file this change does
not touch. CI runners have a writable /tmp.

Fixes #357

[github-actions]

PR metrics

Performance (gcc example, Linux)

Workload	Instructions	CPU time	Page faults	D1 miss	LL miss	Wall	Peak RSS
parse	1693 M (-0.3%)	0.66 s	14.9 k	0.7%	0.1%	0.654 s	33.8 MB (-0.2MB)
dry-run	2308 M (-2.5%)	0.74 s	16.3 k	0.8%	0%	0.745 s	39 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	0 (-100.0%)
Hashes skipped (stat cache)	5833 (+3.5%)
Stat calls	5885
Parse time (ms)	567.1 (+1.6%)
Total time (ms)	728.4 (+1.6%)
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.

Binary size (Linux)

Binary	.text	.data	.bss	File
putup	565.7 KB (+0.1%)	2.3 KB	98.7 KB	671 KB

Code churn (whole codebase, last 30d)

Files	Lines written	Still present	Churned	Churn rate
114	9622	7960	1662	17.3%

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. 4371 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	592
src/graph/scanners/clang_cl.cpp	297
src/graph/scanners/gcc.cpp	116
src/cli/context.cpp	82
src/core/layout.cpp	69
src/graph/scanners/dep_words.cpp	64
src/graph/builder.cpp	62
src/platform/file_io-posix.cpp	42
src/cli/strict_checks.cpp	36
src/cli/cmd_clean.cpp	35

Test coverage (lines)

Overall	Median file	Min file	Max file
88.1%	96.5% (+0.1pp)	14.7% include/pup/parser/token.hpp	100.0% include/pup/core/arena.hpp

102 files · 16608/18845 lines covered

Deltas vs main@26ddb6105.

Updated for 4ff2d32