✨ Include span information in format_result

[elbeno]

Problem:

• When outputting log information to JSON, some information is lost from the
  format_result: namely the types and resulting spans (string locations) of
  the values that were compile-time formatted.

Solution:

• Adjust format_result to include information about all values and their
  locations in the format string.

[lukevalenty]

Is the intention for locations of both runtime and compile-time values to be listed in the spans? For example, spans for compile-time values would show the start and end of the actual compile-time value string representation in the formatted string...and spans for runtime-values would show the start and end of the format specifier "{...}".

ETA: I am indifferent to the result, just curious to know which direction this is going in.

[lukevalenty]

What happens when format results are formatted into another format result? I would expect to see multiple spans.

For example:

auto introduction = format("my name is {}"_cts, "Luke"_cts);
auto result = format("hello, {}"_cts, introduction);

// span 0                       |-------------|
// span 1                                  |--|
static_assert(result == "hello, my name is Luke"_cts);

I would expect a span for each substituted compile-time value at each level.

[elbeno]

• It is the intent to treat runtime and compile-time args mostly the same; in particular to have both types represented in the argument list and the spans list. Makes back-end treatment easier.
• When a format_result is interpolated, the resulting format_result is flattened, not nested. This was already the case and this PR continues that. It's much easier to deal with.