perf: the examples that sweep by hand solve their points side by side, through parallel::map_in_order

[tachsin]

Closes #137 (its jobs' half was #146).

What

• photonoxide::parallel::map_in_order(items, f): f on each item side by side on rayon's threads, the results in the items' order, so a caller that prints or records them afterwards writes what a loop over them writes. A library helper, as the issue asked, rather than threading in each example: the studio builds the examples in and doesn't depend on rayon.
• strip_waveguide (three grids), hadley_corners (four figures × five grids, both schemes) and group_index (five widths, each tracked over the wavelengths) solve their problems with it and then print in order. Their output is byte for byte examples/output/'s (checked for all three).

Measured

photonoxide bench example/... --threads 1,20, Core Ultra 7 265K, same session:

example	1 thread	20 threads, before	20 threads, after
hadley_corners	7.3 s	9.4 s	2.5 s
group_index	10.6 s	13.1 s	3.0 s
strip_waveguide	6.1 s	7.0 s	5.9 s (its 5 nm grid's solve alone)

Each used to be slower on 20 threads than on one. docs/benchmarks.md regenerated; ROADMAP ticks "Sweeps side by side".

Checked

• parallel's test: results in order on 1, 3 and 8 threads; its doctest.
• cargo test --release (361 passed), the studio's tests (35), clippy on both crates, rustdoc with -D warnings.