Remove use of sscanf for reading surface coefficients

[paulromano]

Description

Periodically, users have run into issues (e.g., this recent post) where OpenMC produces the following error:

 ERROR: Something went wrong reading coeffs for surface 1

even though their geometry.xml file looks perfectly fine. I believe this is happening for international users who have a locale set such that the sscanf function (which is currently used for this) is sensitive to it. This PR removes the use of sscanf in favor of C++ streams (implemented in get_node_array).

To make sure this doesn't impact read performance on large geometry.xml files, I tested it on my ITER E-lite model (~500k surfaces) and it made no difference in the initialization time.

Checklist

• I have performed a self-review of my own code
• I have run clang-format on any C++ source files (if applicable)

[gridley]

Hm, while we're here, seems like we should just lump these into one function! Seems like a holdover from the fortran era to have this split into multiple ones. I'll cook up an implementation of this in a bit.

[gridley]

Alright, I just opened a new PR with the variadic version of this. It's up to you whichever way you want merged in the end, Paul!

[paulromano]

@gridley Not a huge fan of the primitive va_list approach so I decided to pursue an alternative here using std::initializer_list. Personally I think it's cleaner and requires less mental effort for someone reading the code. Let me know what you think!

[gridley]

Oooo, good way to do it!!