Switch to vectorized Stokes solver and stop using steady-state itterations

[andrewdnolan]

This PR alters the Stokes solver and thermomechanical coupling strategy, in an attempt to address the persistent failure of some beta value in the 03_PeriodicSurge experiment. There is no clear pattern of simulation failures as a function of beta value too suggest it's a numerical stability problem. The failures manifests through the Stokes solver diverging and the large resulting velocities producing NaN strain heating values.

Per personal communication with Adrien Gilbert, he has made two suggestion, which have been implemented in this PR:

1. Stop using steady-state iterations
2. switch to the vectorized Stokes solver

Upside to both of these changes is they'll dramatically reduce simulation runtimes.

[andrewdnolan]

Steady-State Testing

Running the steady-state initialization experiments with the new solver settings results in:

From trial and error we will need to change the reference glaciers $\Delta\dot{b}$ value from $0.36$ to $0.35$ $\text{m a}^{-1}$

[andrewdnolan]

source ./periodic_surge.sh

parse_params()
{
  IFS=" " read -r QP beta NT T_0<<< $(sed -n "${1}p" "run/${KEY}/${KEY}.commands" | cut -d " " -f 7,9,11,13)
}

parse_runtime(){
    runtime=$(tail logs/${KEY}/${KEY}_52723013_${1}.out | awk '/[S]OLVER TOTAL/ {print $5}')
}

SP=2
KEY='crmpt12'
offset=-0.35
# get the number of simulations to process
N=$(wc -l < "run/${KEY}/${KEY}.commands")

parse_json "params/${KEY}.json"

for j in $(seq $N);do
    parse_params $j
    parse_runtime $j

  log_runtime $KEY $dx $T_ma $offset $ST_dt $SD_dt $QT_dt $QD_dt $SP $QP $TT $beta $runtime
done