Py-grama

*Breaking changes*
- deprecate `tf_transmute()`
- rename `Marginal.l()` to `Marginal.d()`; this is a density, not a likelihood
- rename `marg_named()` to `marg_fit()`; that name is confusing now that we also have `marg_mom()`
- change argument order for `marg_fit()` to make consistent with `marg_mom()`
- rename `colmin` -> `min`, `colmax` -> `max`, `colsum` -> `sum`
-

*Features*
- add random variable sample to marginals `Marginal.r(n)`
- add frozen location parameter option to `marg_mom()` and `marg_fit()`; enables access to 2-parameter lognormal and 2-parameter Weibull distributions
- make marginal summary output more readable by reporting standard moments, rather than distribution parameters
- add autoplot dispatch from `eval_contour()` output
- add `linspace()` and `logspace()` helper functions
- add thermal-radiation channel model based on Andrew Banko's thesis; verify implementation against existing data
- add `tran_iocorr()` to compute input/output correlations; autoplot dispatch for useful EMA tool (correlation tileplot)
- add `mean_lo|up()` and `pr_lo|up()` helper summary functions to compute lower and upper confidence interval bounds
-

*Fixes*
- fixed bug with automatic levels in `eval_contour()`
- correct buckling plate model; now takes wavenumber for buckling mode
- remove dependency on psdr package by inboarding & trimming code, allows us to keep `tran_varproj()` without a cvxpy dependency
- vectorize the cantilever beam model's functions

*Documentation*
- add a more informative example in docstring for `if_else()`
- add documentation to `eval_sample()` that mentions `tran_sp()` for reducing model runtime
- add link to random variable modeling page in documentation to undefined copula error

**Breaking Changes**
- renamed `eval_monte_carlo()` to `eval_sample()`
- changed default argument behavior for `n` in `eval_sample()`; no longer provides default `n=1`

**Infrastructure Changes**
- Moved dependencies from `requirements.txt` to `setup.py`
- Changed optional dependencies to fail at runtime, rather than at import. Also moved optional dependencies to main grama namespace.

**Features**
- Added importance sampling with `tran_reweight`; effective sample size approximation with `neff_is()`
- Added active subspace approximation with `tran_polyridge()`
- Added marginal fitting by the method of moments with `marg_mom()`
- Added fumulative standard deviation window function `cumsd()`
- Added plotnine under-the-hood for visualization; plotnine functions available through Grama namespace for convenience
- Re-implemented autoplot utilities using plotnine
- Added contour generation through marching squares with linear interpolation via `eval_contour()`
- Made `Marginal` class functions `Marginal.l()`, `Marginal.q()`, `Marginal.p()` symbolic to enable use within `tran_mutate()`
- Added QQ plot helper `qqvals()`
- Added grouped counting helper `tran_count()`
- Added `df_group()` helper to generate sweeps

**Fixes**
- Vectorized trajectory model
- Fixed bugs in PRLC model, added unittest coverage

*Additions*:

- Summary functions `gr.skew()` and `gr.kurt()`

*Fixes*:

- Fix behavior of `tran_pivot_longer()` when using `.value` and `names_sep` simultaneously; no longer have to provide dummy `values_to` argument.

Additions:
- `gr.tran_pivot_wider()` and `gr.tran_pivot_longer()` for reshaping data; patterned off of the [pivoting](https://tidyr.tidyverse.org/articles/pivot.html) tools in `tidyr` OscarDeGar

Fixes:
- Docstrings for `dfply` functions (such as `gr.tf_mutate`) fixed; this makes it much easier to reference these tools when coding OscarDeGar
- Eliminated some circular importing by restructuring `core.py` and `tools.py`; adding a new `marginals.py` to consolidate marginal-related tools Riya-1

Thanks to:
- OscarDeGar
- Riya-1

Features:
- Add ability to return p-value for `gr.corr()`
- add `floor()`, `ceil()`, `round()`
- standardize fitting routines to provide a *suffix* on predicted quantities; by default this is `_mean`
- update `tran_kfolds()` to make use of new suffix standard
- added `df_init` kwarg to `eval_nls()` to specify initial parameter guess(es); `fit_nls()` can also access this kwarg
- added `tran_md()` utility; useful in pipelines

Bugfixes:
- `tran_sp()` has issues with strongly anisotropic data; added `standardize` option to solve

New features:
- Support points design of experiments utility: [implemented](https://github.com/zdelrosario/py_grama/blob/master/grama/support.py) in `tran_sp()`; verified in [notebook](https://github.com/zdelrosario/py_grama/blob/master/tests/longrun/sp_convergence.ipynb)
- [Polynomial featurizer](https://github.com/zdelrosario/py_grama/blob/master/grama/tran/tran_scikitlearn.py) via `tran.tran_poly()`
- [Parallel RLC](https://github.com/zdelrosario/py_grama/blob/master/grama/models/circuit_RLC.py) models via `models.make_prlc()` and `models.make_prlc_rand()`
- [Linear model](https://github.com/zdelrosario/py_grama/blob/master/grama/fit/fit_scikitlearn.py) via `fit.fit_lm()`
- [Correlation](https://github.com/zdelrosario/py_grama/blob/master/grama/dfply/summary_functions.py) summary function via `corr()`

Improvements:
- enforced consistency in arguments; particularly `n_maxiter`
- gaussian process implementation modified to provide predictive standard deviation
- ability to subselect outputs for `tran_kfolds()`
- vectorize RLC and cantilever beam models

Bugfixes:
- Fixed silent failure of `marg_named()` and `marg_gkde()` when passing a DataFrame
- Fixed FORM RIA / PMA [implementations](https://github.com/zdelrosario/py_grama/blob/master/grama/eval_tail.py) to return correct MPP