stdlib format improvements - #11502

A negative precision passed via ~.* (e.g. io_lib:format("~.*c", [-1, $a]))
would cause infinite recursion in chars/2 since -1 bsr 1 = -1.
Other controls (~.*f, ~.*s, etc.) would crash with function_clause.

Fix by validating precision >= 0 in precision/2, raising badarg
consistently for all controls. This matches C printf semantics
where negative precision via * is defined behavior (treated as
omitted), but Erlang chooses to reject it explicitly.

The latin1 list clause in flat_trunc/3 used lists:flatten + lists:split
which materializes the entire input (O(L)) just to take N chars.
Replace with string:slice/3 which is O(N), matching the existing
unicode clause. Also merge both list clauses into one since
string:slice handles both encodings.

The ?IND macro called list_to_binary(Ind) on every line wrap in the
binary pretty-printer loops (pp_tail_bin, pp_pairs_tail_bin,
pp_fields_tail_bin, pp_binary_bin_ind). Since Ind is constant within
each loop, this re-flattened the same iolist on every sibling element
that wrapped to a new line.

Fix by converting Ind to binary lazily only in the branch that
actually wraps, then threading the binary through subsequent
iterations. The ?IND macro now expects a binary argument directly.

Previously, chars_limit only capped the field width for ~s via
limit_field/2. Other controls (~c, ~n, ~~, ~w, ~W, ~e, ~f, etc.)
could allocate unbounded padding that bypassed chars_limit entirely.

Fix by passing chars_limit to build_small/build_small_bin and
applying limit_field to the field width before calling control_small.
Also apply limit_field in control_limited for ~w/~W.

Without chars_limit set, behavior is unchanged.