Add special optimization of ExternalSourceAspect and remove some unnecessary code

[khanaffan]

Closes: https://github.com/iTwin/itwinjs-backlog/issues/1479

Changes

• BloomFilter is enabled again. (Need Perf testing if it has any negative effect anymore)
• Disqualifing ECClassId lookup index for ExternalSourceAspect is removed.
• Disqualify JOIN expression for ExternalSourceAspect.Element.Id is added.

This query optimization involves ExternalSourceAspect, which is typically slow. A previously mentioned method was suggested to accelerate its performance.

Reason:

ExternalSourceAspect.Element.Id is associated with the index (ix_bis_ElementMultiAspect_fk_bis_ElementOwnsMultiAspects_target), which closely matches Element.Id. Consequently, using this index for looping yields performance akin to a SCAN operation. Therefore, the index should be used as a last resort, not as a primary or secondary option.

The patch identifies the search expression 'ExternalSourceAspect.Element.Id' and appends a '+' to invalidate it. Some earlier optimizations related to this have been eliminated.

• [ X] Need ECPresentation performance testing as prerequisite to merge. @grigasp

[grigasp]

Need ECPresentation performance testing as prerequisite to merge.

I'll try to find some time for this ASAP, but that'll probably only happen next week.

[grigasp]

Not good... Here're the results:

Test case	4.11	5.0-dev.115	5.0-affan	5.0-affan-analyzed
L iModel 1	10.47	11.20	102.91	93.407
L iModel 2	17.74	19.71	189.73	182.022
L iModel 3	4.00	4.23	86.17	83.358
L iModel 4	21.49	22.39	> 10 min.	> 10 min.
L iModel 5	39.91	42.27	> 10 min.	> 10 min.
L iModel 6	3.14	3.26	4.53	4.464
L iModel 7	1.15	1.35	5.14	4.537
L iModel 8	21.92	23.74	234.33	311.151
M iModel 1	1.02	1.12	1.78	1.403
M iModel 2	2.41	2.57	8.67	7.657
M iModel 3	5.71	23.58	207.52	191.521
M iModel 4	4.02	4.53	96.53	88.335
M iModel 5	1.53	1.92	2.50	2.289
M iModel 6	3.24	3.75	20.27	18.78
M iModel 7	0.33	0.34	0.34	0.326
M iModel 8	0.13	0.14	0.14	0.151
M iModel 9	1.96	1.80	2.08	2.352
M iModel 10	0.22	0.24	0.27	0.256
M iModel 11	0.97	1.01	1.30	1.354
M iModel 12	0.16	0.15	0.18	0.146
M iModel 13	0.37	0.47	1.18	1.167
M iModel 14	0.69	0.65	0.80	0.771
S iModel 1	0.74	0.65	0.66	0.562
S iModel 2	0.27	0.22	0.24	0.212
S iModel 3	0.05	0.05	0.05	0.042
S iModel 4	0.05	0.05	0.05	0.047
S iModel 5	0.79	0.74	0.72	0.678
S iModel 6	0.06	0.05	0.07	0.048
S iModel 7	0.05	0.05	0.05	0.041
XL iModel 1	98.99	93.36	122.13	59.821
XL iModel 2	85.69	88.85	> 10 min.	> 10 min.
XL iModel 3	103.23	118.50	> 10 min.	> 10 min.
XL iModel 4	34.07	49.47	> 10 min.	> 10 min.
XL iModel 5	55.76	79.31	> 10 min.	> 10 min.
XL iModel 6	50.10	72.62	> 10 min.	> 10 min.
XL iModel 7	3.54	4.10	> 10 min.	> 10 min.
L iModel 9	32.99	36.09	> 10 min.	> 10 min.
L iModel 10	7.93	8.38	66.88	24.203
M iModel 15	3.06	3.46	56.83	21.265

This is for a test defined here: https://github.com/iTwin/presentation-performance-tests/blob/d7d9e4db41b2e3b3ac92a636fff3ddfd05614676/src/tests/ContentDescriptor.test.mts#L14. The library runs a number of queries for this request, I haven't investigated further which queries exactly got slower.

[grigasp]

@khanaffan, based on our internal discussion, I re-ran the test on the test iModels after running ANALYZE on them, using the addon built from this branch. The results are better in a few cases, but for the most part they're still much worse than what we have in 4.11 - see updated table above.