fix: the figures carried a chunk WebKit refuses, which is why a page never loaded

The user's report was precise and I misread it twice: opening trigger.page as a *file* rendered the
page with no image, while opening it through the app's `help:` URI spun forever. Both are one cause.
The three figures carried a `bKGD` chunk -- and the PNG specification does not permit `bKGD` on an
image with an alpha channel, which all three are (colour type 6). ImageMagick both writes and reads
it, so my "decodes fully" check proved nothing: I tested with a decoder more permissive than the one
that has to load the file. WebKit rejects it, so the image never appears; over `help:` the fetch
never completes either, so the page's load event never fires and Yelp's spinner never stops.

It came from `import`, which records the root window's background. Every figure now carries only the
chunks a decoder is guaranteed to accept (IHDR/IDAT/IEND), and `tools/make-help-shots.sh` normalises
its output at the end so a re-shoot cannot reintroduce it -- with the reason written where the next
reader will be, since "ImageMagick reads it" is exactly the trap.

`tools/check_help.sh` gains the rule, because the four yelp-check subcommands and the structural
rules are all blind to a figure no decoder will accept: a `bKGD` on colour type 4 or 6 is a finding,
checked directly rather than through a permissive tool. It is a whitelist because a decoder is one.

Verified: the rule fires on the committed figures (colour type 6, bKGD present) and passes on the
fixed ones; the figures still decode and are still 1280x800 / 938x420; check_help.sh is green on the
source and installed sets; make -C build test-help passes.

tools/check_help.sh

@@ -3,9 +3,9 @@
 # SPDX-FileCopyrightText: 2026 Martino Ferrari
 #
 # The manual's check, the way tools/check_ui.py is the .ui files' check: it runs the four
-# yelp-check subcommands that need no network, checks the structural rules below, then renders
-# the whole set to HTML with yelp-build so a page that only fails at the transform step is
-# caught too.
+# yelp-check subcommands that need no network, checks the structural rules below, checks the
+# figures, then renders the whole set to HTML with yelp-build so a page that only fails at the
+# transform step is caught too.
 #
 # `yelp-check validate` is deliberately absent, and it is worth being exact about why: it
 # builds the RelaxNG schema locally with `xsltproc` and `mal-rng.xsl`, but the schema it
@@ -15,11 +15,10 @@
 # What that leaves unverified is *structure*, and that is not a small gap: three pages
 # carrying a `<figure>` whose `<title>` was a sentence and which had no `<desc>` opened in
 # nothing, while links, media, orphans, ids and the HTML render all passed. So the rules this
-# project has actually needed are checked here, with xmllint's XPath and no schema at all.
-# Add a rule when a page *teaches* one -- and do not add one you have not seen broken: a rule
-# that fails a legal page is worse than no rule.
+# project has actually needed are checked here. Add a rule when a page *teaches* one -- and do
+# not add one you have not seen broken: a rule that fails a legal page is worse than no rule.
 #
-# Every expression uses local-name() on purpose. A Mallard page declares
+# Every XPath expression uses local-name() on purpose. A Mallard page declares
 # `xmlns="http://projectmallard.org/1.0/"`, and XPath 1.0 has no default namespace, so
 # `//figure` matches *nothing* and every rule below silently passes. Written naively, this
 # check reported "ok" for the very defect it was written for.
@@ -58,11 +57,51 @@ no text sits directly under page|count(/*[local-name()='page']/text()[normalize-
 a media names its src|count(//*[local-name()='media'][not(@src)])
 RULES
 done
+
+# The figures, and this one is a *decoder's* rule rather than ours: a `bKGD` chunk is forbidden
+# by the PNG specification on an image with an alpha channel, and `import` writes one because it
+# records the root window's background. ImageMagick both writes and reads it, so a round trip
+# through it proves nothing -- which is how three figures passed every check while WebKit
+# refused them, showing a figure that never loaded and, over the `help:` URI, a page that spun
+# forever. Only the chunks PNG allows for the colour type are accepted; a whitelist, because a
+# decoder is one.
+python3 - "$dir" <<'PY' || failed=1
+import glob, os, struct, sys
+seen = 0
+for path in sorted(glob.glob(os.path.join(sys.argv[1], 'figures', '*.png'))):
+    seen += 1
+    with open(path, 'rb') as f:
+        if f.read(8) != b'\x89PNG\r\n\x1a\n':
+            print(f"  {os.path.basename(path)}: not a PNG", file=sys.stderr)
+            continue
+        chunks = []
+        while True:
+            head = f.read(8)
+            if len(head) < 8:
+                break
+            length, kind = struct.unpack('>I4s', head)
+            f.read(length)
+            f.read(4)
+            chunks.append(kind)
+    # Colour type 4 (grey+alpha) and 6 (RGB+alpha) have an alpha channel; bKGD is not permitted
+    # on either. Anything else is left alone: this checks the rule a page was broken by.
+    colour_type = None
+    with open(path, 'rb') as f:
+        f.read(16)
+        colour_type = f.read(1)[0]
+    if colour_type in (4, 6) and b'bKGD' in chunks:
+        print(f"  {os.path.basename(path)}: bKGD on an image with alpha (colour type {colour_type})"
+              " -- PNG forbids it and a strict decoder refuses the file", file=sys.stderr)
+        sys.exit(1)
+if seen == 0:
+    print("  no figures to check", file=sys.stderr)
+PY
+
 if [ "$failed" != 0 ]; then
-  echo "check_help: structural findings above" >&2
+  echo "check_help: findings above" >&2
   exit 1
 fi
-printf 'structure %s: ok\n' "$dir"
+printf 'structure %s: ok; figures: ok\n' "$dir"
 
 mkdir -p "$tmp/html"
 yelp-build html -o "$tmp/html" "$dir"