Port to fvgp 4.8.1 (toward gpCAM 8.4.0)

Bump the fvgp pin to ~=4.8.1 and adapt to its breaking API changes:
- rename constructor args gp2Scale_dask_client->dask_client and
  gp2Scale_linalg_mode->linalg_mode; store them privately to avoid
  fvgp 4.8's new read-only `dask_client` property
- remove calc_inv (folded into linalg_mode="CholInv"/"Inv" upstream)
- rename update_gp_data/tell kwarg gp_rank_n_update->rank_n_update
- update test_gpCAM for renamed fvgp internals (marginal_density->
  marginal_likelihood, .KVlinalg->.kv)

Sync GPOptimizer/fvGPOptimizer docstrings with fvgp 4.8.1 GP/fvGP
(keeping cost_function/logging and optional x_data/y_data), surface
new fvgp 4.8 features (validation metrics, bump/sle_kernel, new
linalg_mode values) in the API docs, and drop the duplicate-autodoc
fvgp_* attribute entries.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Author: MarcusMNoack

docs/source/api/overview.md

@@ -25,6 +25,21 @@ To get to know gpCAM, check out the [examples](../examples/index.md), download t
 repository and look in `./tests`, or visit the project
 [website](https://gpcam.lbl.gov/).
 
+## New in fvgp 4.8 (gpCAM 8.4)
+
+Because the optimizers inherit from `fvgp.GP`, the following fvgp 4.8 additions are
+available directly on any `GPOptimizer`/`fvGPOptimizer` instance and are listed on the
+[gpOptimizer](gpOptimizer.md) / [fvgpOptimizer](fvgpOptimizer.md) pages:
+
+- **Model-validation metrics** — `mae`, `mape`, `msll`, `interval_score`, `mpiw`, and
+  `coverage_curve` for quantifying predictive accuracy and calibration, plus
+  `plot_observed_vs_predicted` for a quick diagnostic plot.
+- **New kernels** — `bump` and `sle_kernel` (in addition to the existing library), all
+  re-exported through `gpcam.kernels` and documented on the [Kernels](kernels.md) page.
+- **Linear-algebra modes** — the `linalg_mode` argument now accepts `CholInv`/`Inv`
+  (the replacement for the removed `calc_inv` option) and preconditioned sparse solvers
+  such as `sparseCGpre`/`sparseMINRESpre` for large gp2Scale problems.
+
 ## See Also
 
 - [Repository](https://github.com/lbl-camera/gpCAM/)

gpcam/autonomous_experimenter.py

@@ -139,20 +139,6 @@ class AutonomousExperimenterGP:
         newly suggested data points will be communicated. The default is False.
     compute_device : str, optional
         One of `cpu` or `gpu`, determines how linear algebra computations are executed. The default is `cpu`.
-    calc_inv : bool, optional
-        If True, the algorithm calculates and stores the inverse of the covariance
-        matrix after each training or update of the dataset or hyperparameters,
-        which makes computing the posterior covariance faster (3-10 times).
-        For larger problems (>2000 data points), the use of inversion should be avoided due
-        to computational instability and costs. The default is
-        False. Note, the training will not use the
-        inverse for stability reasons. Storing the inverse is
-        a good option when the dataset is not too large and the posterior covariance is heavily used.
-        Caution: this option, together with `append=True` in `tell()` will mean that the inverse of
-        the covariance is updated, not recomputed, which can lead to instability.
-        In application where data is appended many times, it is recommended to either turn
-        `calc_inv` off, or to regularly force the recomputation of the inverse via `gp_rank_n_update` in
-        `update_gp_data`.
     training_dask_client : distributed.client.Client, optional
         A Dask Distributed Client instance for distributed training. If None is provided, a new
         `dask.distributed.Client` instance is constructed.
@@ -198,11 +184,10 @@ def __init__(self,
                  x_data=None, y_data=None, noise_variances=None, dataset=None,
                  communicate_full_dataset=False,
                  compute_device="cpu",
-                 calc_inv=False,
                  training_dask_client=None,
                  acq_func_opt_dask_client=None,
                  gp2Scale=False,
-                 gp2Scale_dask_client=None,
+                 dask_client=None,
                  gp2Scale_batch_size=10000,
                  ram_economy=True,
                  args=None
@@ -337,20 +322,6 @@ class AutonomousExperimenterFvGP(AutonomousExperimenterGP):
         newly suggested data points will be communicated. The default is False.
     compute_device : str, optional
         One of `cpu` or `gpu`, determines how linear algebra computations are executed. The default is `cpu`.
-    calc_inv : bool, optional
-        If True, the algorithm calculates and stores the inverse of the covariance
-        matrix after each training or update of the dataset or hyperparameters,
-        which makes computing the posterior covariance faster (3-10 times).
-        For larger problems (>2000 data points), the use of inversion should be avoided due
-        to computational instability and costs. The default is
-        False. Note, the training will not use the
-        inverse for stability reasons. Storing the inverse is
-        a good option when the dataset is not too large and the posterior covariance is heavily used.
-        Caution: this option, together with `append=True` in `tell()` will mean that the inverse of
-        the covariance is updated, not recomputed, which can lead to instability.
-        In application where data is appended many times, it is recommended to either turn
-        `calc_inv` off, or to regularly force the recomputation of the inverse via `gp_rank_n_update` in
-        `update_gp_data`.
     training_dask_client : distributed.client.Client, optional
         A Dask Distributed Client instance for distributed training. If None is provided, a new
         `dask.distributed.Client` instance is constructed.
@@ -396,11 +367,10 @@ def __init__(self,
                  x_data=None, y_data=None, noise_variances=None, dataset=None,
                  communicate_full_dataset=False,
                  compute_device="cpu",
-                 calc_inv=False,
                  training_dask_client=None,
                  acq_func_opt_dask_client=None,
                  gp2Scale=False,
-                 gp2Scale_dask_client=None,
+                 dask_client=None,
                  gp2Scale_batch_size=10000,
                  ram_economy=True,
                  args=None