Minor BJData fixes

[falbrechtskirchinger]

These are the previously discussed minor fixes to the BJData binary_reader code.

• Replace the std::vector/std::map lookup tables with std::arrays.
  This is meant to be an optimization (mostly avoiding unnecessary, repeated heap allocations). Whether std::lower_bound() is actually faster than std::find/std::find_if in this scenario is unknown.

• Replace string_t::npos with binary_reader::npos.

[coveralls]

Coverage remained the same at 100.0% when pulling 98ef5b5 on falbrechtskirchinger:bjdata-fixes into df98ca2 on nlohmann:develop.

[falbrechtskirchinger]

Wow! This improves performance parsing a 136M document by 0.3%. This was truly worth it! 🤣

(It still avoids heap allocations, though.)

[falbrechtskirchinger]

Looks good to me.

[nlohmann]

Looks good to me.