[FIX] scc: don't prefix special characters with a base char, and fix out-of-bounds control code at column 0

[cfsmp3]

Follow-up to #2301. Two defects in the SCC/CCD writer, both found while deep-reviewing that PR.

1. Special characters gained a spurious leading space

#2301 emits a fallback base character before every internal code >= 0x80. That is right for extended characters (0x90-0xcf, hi 0x12/0x13) — handle_extended() decrements cursor_column, so they destructively replace the cell before them and the stream has to carry something for them to overwrite.

It is wrong for special characters (0x80-0x8f, hi 0x11). handle_double() writes those without moving the cursor back, so the base character stays on screen and every special character comes back one column further right.

reference    : ♪ (Wendy Talbot) DEAR JOHN,
master today :  ♪ (Wendy Talbot) DEAR JOHN,   <- spurious space
this PR      : ♪ (Wendy Talbot) DEAR JOHN,

Fix: only emit the base character for extended codes. check_padding() still runs in both cases so the two-byte code starts on an even offset and lands inside a single SCC word.

Testing

Generated an SCC exercising all 80 codes in 0x80-0xcf, re-encoded it and decoded it again:

	wrong of 80
before #2301	79
master (with #2301)	16
this PR	0

Real samples: 725a49f871 (15 music-note rows) and c032183ef0 (3) now round-trip with no altered text. The apostrophe from #2098 is unchanged — still a7 80 92 29.

2. Out-of-bounds control code index when a style change starts at column 0

get_preamble_code() and get_tab_offset_code() take unsigned char. The column - 1 used to place the preamble one cell to the left wraps to 255 at column 0, giving 255 / 4 = 63 and an index far past the end of control_codes[] — row 12 produces code 186 against CONTROL_CODE_MAX 147.

In --out=ccd that entry's assembly pointer is handed to strlen() and segfaults. In --out=scc it silently emits whatever ints happen to follow the array.

This read predates #2301 — 2 of 20 local samples hit it both before and after — but #2301 shifted the binary layout so the garbage pointer now lands in unmapped memory. It went from quietly wrong to a hard crash:

pre-#2301 build : --out=ccd exit 0   (instrumented: 'BADCODE 186, max 147')
master today    : --out=ccd SIGSEGV
this PR         : --out=ccd exit 0

Fix: clamp the preamble column to 0, which is what the adjacent space branch already does.

Regression scope

• Non-SCC output (txt, sami, srt, ttxt, webvtt, g608) byte-identical to master across 20 samples.
• Of those 20, SCC output changes on 12 that contain special characters and 1 that hit the out-of-bounds index; the other 7 are unchanged.
• Valgrind: no invalid reads. Only the pre-existing 32-byte init_encoder leak, which master has too.
• clang-format clean.