Turns out that read without params supports backslash escaping; that is,
`\t` gets translated to tab and `\` at the end of line connects with
next line.  Such feature is dangerous as implicit; read should almost
never be used without this parameter.

See [Bash Hackers wiki]1] for more.

  [1]: http://wiki.bash-hackers.org/commands/builtin/read#read_without_-r

Somehow this did not break anything. The sole fact that the argument was
"ignored" was most bug in itself but we're not going to fix it as the
code gets rewritten any moment now.

Tolerate spaces before the '[' and anything after the ']'.

Note that strict mode has nothing to do with this; we just need to get
the match right.

This reverts commit a6dfb1bfd7008d99b64e4a5bde19fbfc2c66ff10.

Turns out that the behavior suggested in original comment is actually
the opposite from PATH, which is not intuitive, and makes it harder to
support for user customizations with inigrep.inigrep__mkpath, at least
as it's used in saturnin.

The 'caller' and 'frontend' may be hard to grasp; example should help.

Remove iterated debugs (eg. over FFOO_INIGREP_PATH) and those appearing
as arguments

Seems to be more readable and shows ~15% better performance on full
suite run.

So that it's easier to check if we are using the right SUT in case it's
both installed and deployed in temporary test dir.

If user does not provide mandatory parameter, `shift 2` will fail but
will not shift anything, resulting in infinite loop.

Fixes issue when empty string generated warning.  Correct behavior is
that empty string does return with non-zero but does not issue warning
as this is to be commonly used to set default values.

 *  Do not split text per line,
 *  use `<pre/>` for main text,
 *  use `<span />` for debug headers,
 *  and use more suitable class names.

Not sure why I did it in the first place.