Minor updates to kalman performance

[jlperla]

• Fix up benchmarking to get rid of DistributionsAD
• Look for low-hanging fruit of inplace, reusing factorizations, etc.

[jlperla]

The results of he PkgBenchmark judge on the zygote vs. custom gradient prior to optimizing it:

Results

A ratio greater than 1.0 denotes a possible regression (marked with ❌), while a ratio less
than 1.0 denotes a possible improvement (marked with ✅). Only significant results - results
that indicate possible regressions or improvements - are shown below (thus, an empty table means that all
benchmark results remained invariant between builds).

| ID                                          | time ratio                   | memory ratio                 |
|---------------------------------------------|------------------------------|------------------------------|
| `["linear", "FVGQ", "kalman_gradient"]`     | 0.32 (5%) :white_check_mark: | 0.62 (1%) :white_check_mark: |
| `["linear", "rbc", "kalman_gradient"]`      | 0.15 (5%) :white_check_mark: | 0.29 (1%) :white_check_mark: |

[jlperla]

Comparison to old pre-optimized code is now:

Results RELATIVE TO PRE-CUSTOM CODE

ID	time ratio	memory ratio
["linear", "FVGQ", "kalman"]	0.18 (5%) ✅	0.49 (1%) ✅
["linear", "FVGQ", "kalman_gradient"]	0.13 (5%) ✅	0.07 (1%) ✅
["linear", "rbc", "kalman"]	0.10 (5%) ✅	0.64 (1%) ✅
["linear", "rbc", "kalman_gradient"]	0.02 (5%) ✅	0.05 (1%) ✅

Results RELATIVE TO CODE BEFORE ALLOCATIONS

ID	time ratio	memory ratio
["linear", "FVGQ", "kalman"]	1.07 (5%) ❌	1.38 (1%) ❌
["linear", "FVGQ", "kalman_gradient"]	0.41 (5%) ✅	0.12 (1%) ✅
["linear", "rbc", "kalman"]	1.06 (5%) ❌	1.16 (1%) ❌
["linear", "rbc", "kalman_gradient"]	0.12 (5%) ✅	0.17 (1%) ✅

• Note that this is before moving any of the data into a reusable cache between samplers, which should speed things up considerably.