style: Strip stray control characters instead of excluding files

[echoix]

Follow-up to #7459, resolving the discussion in #7888.

The editorconfig-checker bump to v3.11.3 added a pre-decode binary-detection
check that flags a file as binary, and so fails the utf-8 charset check, if it
contains any C0 control byte. Four files were marked charset = unset as a
mechanical workaround. In #7888, @nilason confirmed the bytes in three of them
carry no meaning and are better removed, including the vendored ccmath file,
which is no longer maintained upstream.

Changes

• raster/r.drain/README, scripts/r.drain/README: drop a stray 0x18 (CAN)
  in line 24, that <CAN>differ a little from r.drain.
• lib/external/ccmath/C01-matrix: lines 880-881 held a 0x13/0x16
  (DC3/SYN) pair acting as a subscript toggle around U1. The same matrix is
  written plainly as U1 everywhere else in that file, including at lines
  682-694, 788, 803, 811 and 868, and in the very signature these two lines
  document (u1 = pointer to store of m by n transformation matrix U1), so
  the reading is confirmed rather than inferred.
• .editorconfig: the exception now covers only configure, whose BEL byte is
  a legitimate awk field separator in autoconf-generated code.

No behavioral change; documentation and build-config text only.

Verification

pre-commit run editorconfig-checker --all-files reports zero charset, utf-8
or binary errors across the tree. lib/external/ccmath/C01-matrix is now
decoded and charset-checked without complaint rather than skipped, which is
the positive confirmation that the strip worked; the two READMEs no longer
appear at all.

The only errors in that run are line-ending ones, which are a local
core.autocrlf=true checkout artifact on Windows (git ls-files --eol shows
i/lf w/crlf on untouched files too), not something this PR introduces.

Closes #7888

AI (Claude Opus 5) was used to locate the byte sequences, verify the U1
reading against the rest of the file, and check the result with the hook.