fix: declare bel/rep/tbc -- audited terminfo against 5 real terminals
Systematically compared this file's declared capability set against
xterm-256color, foot, alacritty, kitty, and st-256color (compiling
both sides with tic + infocmp for a fair, normalized comparison) to
find any other implemented-but-undeclared gaps in the same shape as
bce/cr. Found three:
- bel: feed_ground's BEL case is a deliberate no-op (no bell support
in v1), but that's not a reason to leave it undeclared -- sending it
is a real, harmless, correctly-handled action, the same
structurally-correct-but-static tradeoff already made for blink.
Every reference terminal checked declares it.
- rep: run_csi's 'b' case (REP) has repeated the last graphic
character since early in this project; never declared.
- tbc: run_csi's 'g' case handles TBC mode 3 (clear all tab stops)
via clear_all_tab_stops; never declared. A stale test comment had
conflated this with the genuinely-unimplemented modes 1/2 and
asserted tbc should stay absent -- corrected.
Also checked for the reverse direction (declared capabilities the
code doesn't back) and found none. Genuinely unimplemented features
turned up too (CBT/back-tab, the Insert key, shifted special keys,
modified function keys) -- left undeclared correctly, since there's
no behavior for them to under-promise.
Adds bce/cr-style behavior-linking tests for rep and tbc so these
can't silently drift stale again either.