triangle_test: Compute specification test based on transitive triangles
In adzemski/netprobitFE: Network probit with fixed effects

Compute specification test based on transitive triangles

triangle_test(fit_ML, fit_ttest_theta, gammaS = fit_ML$gammaS_hat,
  gammaR = fit_ML$gammaR_hat, theta = fit_ML$theta_hat,
  rho = fit_ML$rho_hat_MLE, rho0 = 0, bs_iterations = 200,
  alpha = 0.1, bias_estimator = "theorem2", bs_inference = "none",
  trimming_bound = 0, bs_trimming_bound = trimming_bound,
  force_robust = FALSE, bs_force_robust = force_robust,
  compute_other_variances = FALSE)

`fit_ML`	returned object from ML estimation (`MLE_stage1`)
`fit_ttest_theta`	fitted `ttest_theta` object
`gammaS`	vector of sender effects (default: estimated value in `fit_ML`)
`gammaR`	vetor of receiver effects (default: estimated value in `fit_ML`)
`theta`	value used for θ (default: estimated value in `fit_ML`)
`rho`	value used for ρ (default: estimated value in `fit_ML`); not needed if both `bias_estimator = "theorem2"` and `bs_inference = "none"`
`rho0`	value of theta under null hypothesis (default: zero/test significance)
`bs_iterations`	number of bootstrap iterations
`alpha`	alpha used to compute bootstrap critical values
`bias_estimator`	specify estimator of bias term `"theorem1"` based on theorem in paper, needs estimate of rho `"theorem2"` based on theorem in paper, doesn't need estimate of rho `"boostrap"` boostrap the bias
`bs_inference`	specify what kind of bootstrap is used to determine whether the test rejects `"none"` no bootstrap inference `"double"` double bootstrap
`trimming_bound`	specify trimming of probabilities used to compute statistics
`bs_trimming_bound`	specify trimming of probabilities used in simulation of bootstrap distribution
`force_robust`	logical, don't use speedglm
`bs_force_robust`	logical, don't use speedglm in simulation of bootstrap distribution
`compute_other_variances`	logical, force computation of standard errors for infeasible triangle tests