[Feature] Fast matrix assembly

[bodhinandach]

Describe the PR
This PR modifies the assembly procedure of sparse matrices from direct memory access to tripletList. TripletList is conceptually a vector of Eigen::Triplet<double> which contains 2 indices and a double.

Related Issues/PRs
#4, #5

Additional context
This construction can improve the simulation time 1.5 times faster in ImplicitLinear solver. Though I need to run some benchmarks for other multi-phase solvers.

[bodhinandach]

@RyotaHashimoto Can you help me to confirm if this is faster compare to the current code?

[RyotaHashimoto]

The changes look good to me.
I checked the performance of the code for the 3D beam problem using the linear shape function (2 MPI ranks and 8 threads are used) as well as the code review.
The older code took about 42000ms, and the new code took about 24000ms.
So the new code was 1.75 times faster than the old one.
Great job, @bodhinandach !

[bodhinandach]

Thanks for confirming @RyotaHashimoto. Please don't merge this branch as I am still checking the performance of other solvers.

[codecov-commenter]

Codecov Report

Merging #16 (0888d17) into master (d31f823) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master      #16   +/-   ##
=======================================
  Coverage   95.23%   95.23%           
=======================================
  Files         185      185           
  Lines       36713    36716    +3     
=======================================
+ Hits        34960    34963    +3     
  Misses       1753     1753           

Impacted Files	Coverage Δ
...ers/assemblers/assembler_eigen_implicit_linear.tcc	78.48% <100.00%> (+0.85%)	⬆️
include/solvers/mpm_implicit_linear.tcc	83.64% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update d31f823...0888d17. Read the comment docs.

[bodhinandach]

Let's merge this. The Navier-Stokes and Two-Phase assemblers do not work better with this modification. Maybe there is a size effect of the matrix assembly.