Speed up the airfoil drag calculations

[askprash]

This PR makes some improvements to the airfoil drag calculations to speed up the airsection calculations by roughly ~5x. The airfoil type wasn't really type stable/ it's field types weren't inferable by the compiler. Also the spline evaluations have been cleaned up a little and optimized for better performance.

On my M2 mac the performance of the airfoil section calculations are:

julia> bench = @benchmarkable aerodynamics.airfun($clp, $toc, $Mperp, 
                                               $airfoil_section) seconds = 30 evals = 100
Benchmark(evals=100, seconds=30.0, samples=10000)

julia> bench_airfun = run(bench)
BenchmarkTools.Trial: 10000 samples with 100 evaluations.
 Range (min … max):  696.250 ns …  1.106 μs  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     701.670 ns              ┊ GC (median):    0.00%
 Time  (mean ± σ):   710.746 ns ± 24.497 ns  ┊ GC (mean ± σ):  0.00% ± 0.00%

  ▇█▇▆▅▅▄▂▂▂▁▁    ▃▄▃▃▂▁                                       ▂
  █████████████▇▇██████████▆▇▇▆▆▅▆▆▆▇▅▇▆▆▆▅▆▄▆▅▆▅▄▅▄▃▅▄▄▄▅▃▄▂▅ █
  696 ns        Histogram: log(frequency) by time       824 ns <

 Memory estimate: 96 bytes, allocs estimate: 4.

This PR:

julia> bench_airfun = run(bench)
BenchmarkTools.Trial: 10000 samples with 100 evaluations.
 Range (min … max):  131.250 ns … 392.500 ns  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     133.750 ns               ┊ GC (median):    0.00%
 Time  (mean ± σ):   134.427 ns ±   7.138 ns  ┊ GC (mean ± σ):  0.00% ± 0.00%

              ▂  ▆  █  ▆  ▄                                      
  ▂▁▁▃▁▁▃▁▁▆▁▁█▁▁█▁▁█▁▁█▁▁█▁▁█▁▁▅▁▁▄▁▁▃▁▁▂▁▁▂▁▁▂▁▁▂▁▁▂▁▁▂▁▁▂▁▁▂ ▃
  131 ns           Histogram: frequency by time          140 ns <

 Memory estimate: 0 bytes, allocs estimate: 0.

[codecov]

Codecov Report

Attention: Patch coverage is 5.00000% with 57 lines in your changes missing coverage. Please review.

Project coverage is 5.68%. Comparing base (e182855) to head (6856d44).

Files with missing lines	Patch %	Lines
src/aero/airfun.jl	0.00%	31 Missing ⚠️
src/aero/trefftz.jl	0.00%	14 Missing ⚠️
src/aero/surfcd.jl	0.00%	8 Missing ⚠️
src/aero/airfoil.jl	20.00%	4 Missing ⚠️

❗ There is a different number of reports uploaded between BASE (e182855) and HEAD (6856d44). Click for more details.

HEAD has 2 uploads less than BASE

Flag	BASE (e182855)	HEAD (6856d44)
	3	1

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #118       +/-   ##
==========================================
- Coverage   79.14%   5.68%   -73.47%     
==========================================
  Files          82      78        -4     
  Lines       13585   13552       -33     
==========================================
- Hits        10752     770     -9982     
- Misses       2833   12782     +9949     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[ngomezve]

The PR looks good to me. The aerodynamic functions in airfun.jl in general are the least readable in all of TASOPT: they have essentially no comments or documentation and the variable names are not descriptive. Since the PR is not any worse in this regard than main, I believe it can be pulled in.

[ngomezve]

It would be good to comment on why the integer division is used

[askprash]

Got rid of a few more stray allocations that weren't needed. Wing cd calculations are allocation free now and significantly faster relative to main:

julia> bench_surfcd2 = run(bench)
BenchmarkTools.Trial: 10000 samples with 5 evaluations.
 Range (min … max):  8.225 μs …  1.750 ms  ┊ GC (min … max): 0.00% … 99.12%
 Time  (median):     8.542 μs              ┊ GC (median):    0.00%
 Time  (mean ± σ):   9.639 μs ± 29.016 μs  ┊ GC (mean ± σ):  7.69% ±  2.78%

  ▄▇█▇▅▅▄▃▄▄▃▃▂▂▂▂▁▁▁▁▁▁                                     ▂
  ██████████████████████████▇▇▇▆▅▅▆▅▄▆▆▄▅▅▄▄▅▁▅▄▅▁▄▄▄▅▃▄▄▄▃▄ █
  8.22 μs      Histogram: log(frequency) by time     14.6 μs <

 Memory estimate: 8.31 KiB, allocs estimate: 281.

This PR:

---------------------------------------
surfcd2 (FORTRAN on MacPro M2 ~ 2.01 μs)
---------------------------------------
BenchmarkTools.Trial: 10000 samples with 5 evaluations.
 Range (min … max):  1.633 μs …  12.058 μs  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     1.750 μs               ┊ GC (median):    0.00%
 Time  (mean ± σ):   1.782 μs ± 342.426 ns  ┊ GC (mean ± σ):  0.00% ± 0.00%

   ▂▂▂▄█                                                       
  ▃█████▇▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▁▂▂▂▂ ▃
  1.63 μs         Histogram: frequency by time        3.18 μs <

 Memory estimate: 0 bytes, allocs estimate: 0. 

New documentation added too.

[askprash]

Found a bug 🪲. The spline function had an error here:

TASOPT.jl/src/utils/spline.jl

Line 36 in 277bc6d

B[i-1] = ΔS₊