fix(wing): client filter handles multi-word/\s/\b patterns, homoglyphs, marks, fillers; checked in bounded chunks (review of #211, pass 2)

Views of each chunk: as is; NFKD + marks/format/fillers dropped + Cyrillic/Greek
lookalikes mapped + dashes normalised, with whitespace collapsed (acme corp,
\s) or removed (acme-\nclient); and a soft hyphen read as a dash. Chunks of 1 MiB
with a 4 KiB overlap bound the folding's allocation (it was ~14x the input: 919
MiB for a 64 MiB note). Documented limits: leetspeak, punctuation splits,
encodings and homoglyphs outside a small table. Mutation-checked.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

ingestion/internal/wing/router_test.go

@@ -3,6 +3,7 @@ package wing_test
 import (
 	"context"
 	"regexp"
+	"strings"
 	"testing"
 
 	"github.com/stretchr/testify/assert"
@@ -148,7 +149,7 @@ func TestRouter_FolderNamesAreAsciiAndNotReserved(t *testing.T) {
 		dir  string
 	}{
 		{"Kelvin sign lowercases to an ASCII k", "Kubernetes"},
-		{"cyrillic lookalike", "hоmelab"},
+		{"cyrillic lookalike", "h\u043emelab"},
 		{"retired layout name: sources", "sources"},
 		{"retired layout name: concepts", "concepts"},
 		{"retired layout name: entities", "entities"},
@@ -172,3 +173,64 @@ func TestNewRouter_AReservedDefaultIsRefused(t *testing.T) {
 	_, err := wing.NewRouter(routerBrain(t), "sources", "")
 	assert.Error(t, err)
 }
+
+// From the second review of #211: the multi-word, \s and \b patterns an operator
+// would actually write, and the next tier of evasions.
+func TestRouter_ClientMarkerHandlesRealisticPatternsAndHarderEvasions(t *testing.T) {
+	tests := []struct {
+		name, pattern, text string
+	}{
+		{"space in pattern vs NBSP", `acme corp`, "about acme\u00a0corp today"},
+		{"space in pattern vs two spaces", `acme corp`, "about acme  corp today"},
+		{"space in pattern vs newline", `acme corp`, "about acme\ncorp today"},
+		{"backslash-s pattern vs NBSP", `acme\s+corp`, "about acme\u00a0corp today"},
+		{"word boundary vs a zero-width char inside", `\bacme\b`, "the ac\u200bme corp"},
+		{"dash pattern vs a soft hyphen as the dash", `acme-client`, "acme\u00adclient"},
+		{"cyrillic a", `acme-client`, "\u0430cme-client"},
+		{"cyrillic e", `acme-client`, "acm\u0435-client"},
+		{"cyrillic o and c", `coop`, "c\u043e\u043ep"},
+		{"combining acute", `acme-client`, "a\u0301cme-client"},
+		{"precomposed accent", `acme-client`, "\u00e1cme-client"},
+		{"hangul filler", `acme-client`, "ac\u3164me-client"},
+		{"halfwidth hangul filler", `acme-client`, "ac\uffa0me-client"},
+		{"braille blank", `acme-client`, "ac\u2800me-client"},
+		{"grapheme joiner", `acme-client`, "ac\u034fme-client"},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			r, err := wing.NewRouter(routerBrain(t), "web-clip", tt.pattern)
+			assert.NoError(t, err)
+			assert.Equal(t, wing.Exception, r.Resolve(context.Background(), wing.Note{Text: tt.text}).Source, "%q should be caught by %q", tt.text, tt.pattern)
+		})
+	}
+	t.Run("ordinary text is not caught by a multi-word pattern", func(t *testing.T) {
+		r, _ := wing.NewRouter(routerBrain(t), "web-clip", `acme corp`)
+		assert.Equal(t, wing.Default, r.Resolve(context.Background(), wing.Note{Text: "acme is not corp, and corp is not acme"}).Source)
+	})
+}
+
+// The folding allocates several times the text, so a large note is checked in
+// bounded chunks. A marker that straddles a chunk edge must still be found.
+func TestRouter_ClientMarkerIsFoundAcrossAChunkBoundary(t *testing.T) {
+	r, err := wing.NewRouter(routerBrain(t), "web-clip", `acme-client`)
+	assert.NoError(t, err)
+	r.FilterChunk = 64
+	for _, at := range []int{0, 30, 55, 58, 62, 64, 70, 120, 190} {
+		text := strings.Repeat("x", at) + "acme-client" + strings.Repeat("y", 200)
+		assert.Equal(t, wing.Exception, r.Resolve(context.Background(), wing.Note{Text: text}).Source, "marker at offset %d was missed", at)
+	}
+	clean := strings.Repeat("lorem ipsum ", 100)
+	assert.Equal(t, wing.Default, r.Resolve(context.Background(), wing.Note{Text: clean}).Source)
+}
+
+func TestRouter_ClientMarkerChunkingDoesNotSplitAMultiByteRune(t *testing.T) {
+	r, err := wing.NewRouter(routerBrain(t), "web-clip", `acme-client`)
+	assert.NoError(t, err)
+	r.FilterChunk = 65 // odd, over two-byte runes: every boundary falls inside a rune unless it is moved
+	for pad := 0; pad < 8; pad++ {
+		text := strings.Repeat("\u00e9", 40+pad) + "acme-client" + strings.Repeat("\u00e9", 40)
+		assert.Equal(t, wing.Exception, r.Resolve(context.Background(), wing.Note{Text: text}).Source, "pad %d", pad)
+	}
+	clean := strings.Repeat("\u00e9", 300)
+	assert.Equal(t, wing.Default, r.Resolve(context.Background(), wing.Note{Text: clean}).Source, "no panic or false match on a text of 2-byte runes")
+}