st: add bulk ASCII fast path to twrite

Streaming text spent ~50% of its cycles in per-character tputc/tsetchar
dispatch. tputcrun() writes runs of plain printable ASCII directly into
the current line as whole cells, bailing out to tputc() whenever anything
non-trivial is in play (escape state, print/insert/sixel modes, GRAPHIC0
charset, pending wrap, active selection, non-ASCII bytes).

Cuts user time ~3x (0.79s -> 0.26s) and wall time ~45% when streaming
54MB of text. Verified pixel-identical rendering against the previous
build for escape-sequence torture tests, wide/CJK/emoji glyphs,
box-drawing, alt screen, scroll regions, insert/delete, resize reflow,
X primary selection contents and a syntax-highlighted vim session.

st.c

@@ -3707,6 +3707,84 @@ check_control_code:
 	}
 }
 
+/*
+ * Print a run of plain printable ASCII characters using the current cursor
+ * attributes, writing whole cells directly into the current line. Returns the
+ * number of bytes consumed, or 0 when the fast path does not apply and the
+ * caller must fall back to tputc().
+ */
+static int
+tputcrun(const char *buf, int buflen)
+{
+	Glyph attr = term.c.attr;
+	Glyph *gp;
+	int i, run, avail;
+
+	/* Only handle the simple, overwhelmingly common case. */
+	if (term.esc || (term.mode & (MODE_PRINT|MODE_INSERT|MODE_SIXEL)) ||
+	    term.trantbl[term.charset] == CS_GRAPHIC0 ||
+	    (term.c.state & CURSOR_WRAPNEXT) ||
+	    sel.mode != SEL_IDLE || sel.ob.x != -1)
+		return 0;
+
+	avail = term.col - term.c.x;
+	if (avail <= 0)
+		return 0;
+
+	/* Count leading printable ASCII bytes, bounded by the line width. */
+	for (run = 0; run < buflen && run < avail; run++) {
+		unsigned char c = (unsigned char)buf[run];
+		if (c < 0x20 || c >= 0x7f)
+			break;
+	}
+	if (run < 2)
+		return 0;  /* not worth it; keep tputc's semantics */
+
+	attr.mode |= ATTR_SET;
+	attr.mode &= ~(ATTR_WIDE|ATTR_WDUMMY|ATTR_BOXDRAW|ATTR_KITTYPH);
+	#if OSC7_OSC8_PATCH
+	attr.link = curlink;
+	#endif // OSC7_OSC8_PATCH
+	#if KITTY_GRAPHICS_PATCH
+	attr.kitty = 0;
+	#endif // KITTY_GRAPHICS_PATCH
+
+	gp = &term.line[term.c.y][term.c.x];
+
+	/* A wide glyph may straddle the start of the run. */
+	if (gp->mode & ATTR_WDUMMY) {
+		if (term.c.x == 0)
+			return 0;
+		gp[-1].u = ' ';
+		gp[-1].mode &= ~ATTR_WIDE;
+	}
+	/* ...or the end of it. */
+	if (run < avail && (gp[run].mode & ATTR_WDUMMY)) {
+		gp[run].u = ' ';
+		gp[run].mode &= ~ATTR_WDUMMY;
+	}
+
+	for (i = 0; i < run; i++) {
+		gp[i] = attr;
+		gp[i].u = (unsigned char)buf[i];
+	}
+
+	term.dirty[term.c.y] = 1;
+	term.lastc = (unsigned char)buf[run-1];
+
+	if (term.c.x + run < term.col) {
+		term.c.x += run;
+	} else {
+		term.c.x = term.col - 1;
+		#if REFLOW_PATCH
+		term.wrapcwidth[IS_SET(MODE_ALTSCREEN)] = 1;
+		#endif // REFLOW_PATCH
+		term.c.state |= CURSOR_WRAPNEXT;
+	}
+
+	return run;
+}
+
 int
 twrite(const char *buf, int buflen, int show_ctrl)
 {
@@ -3720,6 +3798,14 @@ twrite(const char *buf, int buflen, int show_ctrl)
 	#endif // SYNC_PATCH
 
 	for (n = 0; n < buflen; n += charsize) {
+		/*
+		 * Fast path: a run of plain printable ASCII printed with the
+		 * current attributes onto the current line. This skips the
+		 * per-character tputc/tsetchar dispatch, which dominates the
+		 * profile while streaming large amounts of text.
+		 */
+		if (!show_ctrl && (charsize = tputcrun(buf + n, buflen - n)) > 0)
+			continue;
 		#if SIXEL_PATCH
 		if (IS_SET(MODE_SIXEL) && sixel_st.state != PS_ESC) {
 			charsize = sixel_parser_parse(&sixel_st, (const unsigned char*)buf + n, buflen - n);