`OutputVar` Identifiers, and easy per-variable API for CovarEstimator

## Issue
Currently, `OutputVar`s contain identifiers (e.g. variable names) but once stacked in a `SampleCollection` this ID is lost, therefore, when we draw functions that have dependence on the `OutputVar`(OV), say `CovarEstimator`, we must pass in information that depends on ordering of OV in the sample collection, e.g.
```julia
CovarEstimator(
vec_of_modifiers = v1, # e.g. whether to use latitude weightings
vec_of_other_modifiers =v2, # e.g. a scaling per `OutputVar`
sample_modifier = 2.0, # const applied to all `OutputVar`s in the `SampleCollection`
)
```
This is ripe for confusion

## Possible Solution

1. Impose that identifiers/names must exist and be unique for rows in a sample-collection. A first convention might be `variable_(spatial statistic)_(time statistic)` (for example, `T_field_season`, `T_field_month`,`T_global_month`, `rh_global_season`)
2. Then API for per-variable scalings can be provided per-output-var id;
```julia
per_var_modifier = Dict(
T_field_season => (mod1,mod2), # define modifiers per ID
T_field_month => (mod1,mod2),
T_global_month => (mod1, mod2),
rh_global_season => (mod1,mod2),
)

CovarEstimator(
per_var_modifier = per_var_modifier,
sample_modifier = 2.0
)
```
Note this opens the door in future to more complex modifiers too

Start by tracing OutputVar identifiers through SampleCollection and the CovarEstimator entry point; the issue names no files or tests. Determine how identifiers are currently stored or discarded, then clarify the naming, uniqueness, and per-variable modifier API before implementing it, with completion defined by working identifier-based covariance configuration and corresponding tests.

Written by the indexing model from the issue text.