[analyzer] Various optimizations

In total this shaves off a little over 900 mio instructions when
analyzing the CFE.

Changes:
* informative_data: _InfoBuilder._buildInstanceData - only go though members once
* informative_data: _InfoBuilder.build - only go though unit.directives once
* informative_data: _InfoBuilder.build - only go though unit.declarations once
* inheritance_manager3: Inline first part of _combineSignatures
* informative_data: Don't call withOriginDeclaration several times; don't turn iterators into lists to use them as iterators
* type_visitor: Don't create lists just to visit the content
* inheritance_manager3: Common cases of no substitutions and not covariance optimized; create a few less mostly unused lists
* element.dart: Don't create list to take a specific element

Detailed runtime breakdowns:

Runs of "pkg/analyzer/tool/stable_analysis.aot" with --no-fine-dependencies --no-lints --no-warnings --no-comments --silent run on the golem version of the cfe

Normal GC, 10 runs:

```
msec task-clock:u: -6.6412% +/- 1.1877% (-571.79 +/- 102.26) (8609.84 -> 8038.05)
page-faults:u: -0.5690% +/- 0.0662% (-907.90 +/- 105.67) (159559.80 -> 158651.90)
cycles:u: -6.9461% +/- 1.2300% (-2509675716.90 +/- 444394620.46) (36130860072.10 -> 33621184355.20)
instructions:u: -6.7293% +/- 0.0019% (-2944479732.50 +/- 814962.56) (43756250678.90 -> 40811770946.40)
branch-misses:u: -6.3244% +/- 5.6672% (-9863445.60 +/- 8838488.16) (155958035.20 -> 146094589.60)
seconds time elapsed: -6.6376% +/- 1.1880% (-0.57 +/- 0.10) (8.61 -> 8.04)
seconds user: -6.7709% +/- 1.2279% (-0.56 +/- 0.10) (8.30 -> 7.74)
maxRssKbytes: 0.7329% +/- 0.0215% (3698.00 +/- 108.58) (504581.60 -> 508279.60)
maxRssBytes: 0.7329% +/- 0.0215% (3786752.00 +/- 111181.19) (516691558.40 -> 520478310.40)

Comparing GC data:
Scavenge(   new space) goes from 135 to 131
MarkSweep(   promotion) goes from 16 to 14
MarkSweep(   old space) goes from 0 to 1
Notice combined GC time goes from 3070 ms to 2734 ms (notice only 1 run each).
```

GC "disabled", 10 runs:

```
msec task-clock:u: -5.0663% +/- 1.2758% (-586.83 +/- 147.78) (11583.04 -> 10996.20)
page-faults:u: -3.0945% +/- 0.0002% (-29796.30 +/- 1.81) (962877.60 -> 933081.30)
cycles:u: -5.5270% +/- 1.4680% (-2370772292.80 +/- 629706499.59) (42894735346.20 -> 40523963053.40)
instructions:u: -3.3051% +/- 0.0018% (-916691160.20 +/- 511613.60) (27735747067.70 -> 26819055907.50)
branch-misses:u: -5.8178% +/- 5.3891% (-5252006.40 +/- 4864978.91) (90275210.60 -> 85023204.20)
seconds time elapsed: -5.0598% +/- 1.2754% (-0.59 +/- 0.15) (11.59 -> 11.00)
seconds user: -5.6044% +/- 1.4672% (-0.56 +/- 0.15) (10.02 -> 9.46)
maxRssKbytes: -3.0862% +/- 0.0027% (-119263.20 +/- 103.41) (3864409.20 -> 3745146.00)
maxRssBytes: -3.0862% +/- 0.0027% (-122125516.80 +/- 105886.94) (3957155020.80 -> 3835029504.00)
```

The take-away - for me - is this:
On this compile there's executed about 916 mio instructions less and a reduction of "garbage" created of about 116 MB - all without GC.
With normal GC stuff also happens to the GC (e.g. because of the ~116 MB less "garbage" being created) and it makes the numbers different - in this case both instructions, cycles and time is in the 6-7% savings range (compared to the ~3% instructions saving without GC) - but how much of the extra 3-4 %-points is "real" (because of different GC etc) and how much is "artifact" is a question.


Runs of "pkg/analyzer/tool/stable_analysis.aot" with no extra arguments run on the golem version of the cfe

Normal GC, 10 runs:

```
msec task-clock:u: -1.4721% +/- 0.9488% (-172.44 +/- 111.13) (11713.69 -> 11541.25)
page-faults:u: 0.3663% +/- 0.1061% (706.10 +/- 204.49) (192778.20 -> 193484.30)
cycles:u: -1.5475% +/- 0.9836% (-764366372.80 +/- 485822682.09) (49393259712.10 -> 48628893339.30)
instructions:u: -2.0415% +/- 0.0042% (-1242276662.60 +/- 2581212.09) (60849746427.20 -> 59607469764.60)
seconds time elapsed: -1.4702% +/- 0.9494% (-0.17 +/- 0.11) (11.72 -> 11.55)
seconds user: -1.2812% +/- 1.0658% (-0.15 +/- 0.12) (11.34 -> 11.19)
seconds sys: -7.1783% +/- 7.0964% (-0.03 +/- 0.03) (0.38 -> 0.35)
maxRssKbytes: 0.3406% +/- 0.0199% (2104.00 +/- 123.03) (617689.60 -> 619793.60)
maxRssBytes: 0.3406% +/- 0.0199% (2154496.00 +/- 125986.42) (632514150.40 -> 634668646.40)

Comparing GC data:
Scavenge(   new space) goes from 196 to 192
MarkSweep(   promotion) goes from 16 to 18
MarkSweep(   old space) goes from 1 to 0
Notice combined GC time goes from 3588 ms to 3590 ms (notice only 1 run each).
```

GC "disabled", 10 runs:

```
msec task-clock:u: -3.2974% +/- 0.4563% (-731.49 +/- 101.23) (22184.14 -> 21452.64)
page-faults:u: -2.1323% +/- 0.0001% (-30583.00 +/- 1.74) (1434269.00 -> 1403686.00)
cycles:u: -3.4584% +/- 0.5315% (-2959866681.50 +/- 454870221.35) (85585124094.10 -> 82625257412.60)
instructions:u: -2.1807% +/- 0.0060% (-953640339.30 +/- 2623134.87) (43731091459.60 -> 42777451120.30)
seconds time elapsed: -3.2897% +/- 0.4559% (-0.73 +/- 0.10) (22.20 -> 21.47)
seconds user: -3.2631% +/- 0.6243% (-0.65 +/- 0.12) (19.88 -> 19.23)
seconds sys: -3.5965% +/- 3.3140% (-0.08 +/- 0.08) (2.31 -> 2.22)
maxRssKbytes: -2.1260% +/- 0.0017% (-122282.00 +/- 99.83) (5751738.80 -> 5629456.80)
maxRssBytes: -2.1260% +/- 0.0017% (-125216768.00 +/- 102222.23) (5889780531.20 -> 5764563763.20)
```

The take-away - for me - is this:
This shaves off about 953 mio instructions and ~119 MB of "garbage" without GC.
With normal GC stuff is pushed differently and the change looks smaller than before - about 1.5-2% instead.

Change-Id: I4dc8570717ac21c84b6645c33e78cee364823822
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/479900
Commit-Queue: Jens Johansen <jensj@google.com>
Reviewed-by: Johnni Winther <johnniwinther@google.com>