Analysis - DamageNode optimisations

Split out from #574

* We check for not variant class variant_class" = 'SN' - which makes an inner query - could try and optimise
* We check for Consequence::text like splice or splice regions is not null - we preprocess this?
* We check if any of 6 splice scores are above a threshold - we could preprocess and store max_score for all of them and only check that

--------

**variant class != SN**

* Original inner query - 4.3s
* Tried ref/alt length > 1 - 4.2s
* List indel variant classes (same result but not sure this is exactly the same as there is also ALU insert) - 3.7
* __ne not equal lookup - 3.8s

For the other stuff, since it requires a schema migration, I might do it along with the split up table work

```
# These splicing fields are pre-calculated to make DamageNode queries faster
max_splice_score = models.FloatField(null=True, blank=True) # Of spliceAI and dbscsnv_ada_score/dbscsnv_rf_score
# (Q(consequence__contains='splice')|Q(splice_region__isnull=False)) & Q(variantannotation__variant_class__ne="SN")
splice_indel = models.BooleanField(default=False, blank=True)
```

The only reference to in SA Path prod is in a Qase test case

```
FilterNodeItem.objects.filter(field__icontains='spliceai').values("filter_node__analysis")
```

http://frgeneseq02.imvs.sa.gov.au:92/analysis/18907/

The issue names FilterNodeItem, the Qase lookup, and the proposed max_splice_score and splice_indel model fields; begin by tracing those references and the existing DamageNode filters. Measure the listed variant-class and splice predicates before deciding whether preprocessing or a schema migration belongs with the split-up table work. Done means the agreed query path is faster without changing results, with the relevant migration and checks covered.

Written by the indexing model from the issue text.