re-write flutter analyze (the single-shot and --flutter-repo) to use the analysis server

flutter/engine@9859af1 Roll dart to 29d6e963702fcbb0ed0d5ca67d6b5f977ac0c328

Revert "re-write flutter analyze (the single-shot and --flutter-repo) to use the analysis server (#16281)" (#16482)

This reverts commit 2f41ea54.

re-write flutter analyze (the single-shot and --flutter-repo) to use the analysis server

See https://github.com/dart-lang/sdk/issues/28463, which I think is a
regression.

This also fixes the regression introduced by
https://codereview.chromium.org/2559773002 whereby we were no longer
checking any of the lints.

* Update to Dart SDK `1.22.0-dev.6.0`.

* Update SDK summary generation to create spec and strong summaries.

* Updated to dev.8.0.

* Fixed console out.

This reverts commit ef25a203 to unbreak the build.

* Bump Dart SDK to 1.22.0-dev.5.0.

* Suppress spurious `FutureOr` type warning.

* Fixed linter dep post merge.

* Formatting tweaks.

This ensures that accidental usages of dart:io's
file API don't creep in over time.

This removes direct file access from within flutter_tools
in favor of using `package:file` via a `FileSystem` that's
accessed via the `ApplicationContext`.

This lays the groundwork for us to be able to easily swap
out the underlying file system when running Flutter tools,
which will be used to provide a record/replay file system,
analogous to what we have for process invocations.

Bump to latest Dart SDK dev build (`1.21.0-dev.0.0`).

* updates to analyzer w/ support for bazel workspaces
* removes deprecated analysis `cacheSize` variable access

* Bump Dart SDK to `1.20.0-dev.10.0`.

`1.20.0-dev.10.0` corresponds to `1.20-RC0`.

* pubspec.lock cleanup

* check for pubspec.lock existence

* cleaned up conditional remove

* Roll to Dart SDK `1.20.0-dev.5.0`.

* bumps to `1.20.0-dev.5.0`.
* updates `analyzer` API use.

* Add explicit types.

* Udpate to Dart `1.20.0-dev.1.0`.

Udpate to Dart `1.20.0-dev.1.0` and corresponding `analyzer` package.

* Added `--no-packages-dir`.

* Revert extension to master.

* Added ignores.

* Review nits.

* More nits.

* brings in analyzer version (`0.27.4-alpha.14`) corresponding to current Dart SDK (`1.18.0-dev.2.0`).
* updates analysis to use prefered API for embedder URI resolution
* adds trampolines to `State` and `StatelessWidget` to allow for warning-free within-library @protected access (needed since we closed off access to @protected closures from outside subclasses).
* turns off cache dependency tracking for analysis (in DDC this amounted to a 10% speed improvement).

Cleans up 3 of the 4 violations of the `overriden_field` lint.

The last one is more interesting and I'll defer to someone closer to that code:

    [lint] Do not override fields. (packages/flutter/test/rendering/rendering_tester.dart, line 91, col 14)

Net-net: SDK summaries are not safe in the presence of SDK library embedders.

Context (and bug to track re-enablement) here: https://github.com/dart-lang/sdk/issues/26467.

The latest dev build has stable summaries so we should start using them.

(Also ensures that analysis options are propogating to the SDK analysis context --- see:  https://github.com/dart-lang/sdk/issues/26129.)

Package mapping is already done by the `PackageDependencyTracker` so this extra check is at best not needed.  (At worst could cause an unneeded and costly call to `pub list-package-dirs`!)

Updates the analyze command to pass a package map to analysis rather than a file path.

This allows us to avoid creating a needless temporary `.packages` file and host directory and saves us a trip to disk to retrieve the contents when building our URI resolvers for analysis.

Will have the result of restoring `[static warning]` to `[warning]`, etc.

(This is essentially how we handle it in the analyzer CLI.)

Introduces a new Dart analysis wrapper that works directly with the analyzer API (in favor of shelling out to a separate process).

Some consequences:

  * we no longer need to fear parts (simplifying our dart file gathering)
  * we can filter by error code (when needed), rather than by error strings
  * no more IO scraping
  * no need to generate `main()` or to run with `--package-warnings`
  * we now specify an analyzer (and linter) version in the pubspec (we’ll want to make sure this doesn’t diverge too far from the analyzer shipped with the SDK but it does give us some room to play with experimental builds)
  * no more (re)scanning of error source files (and so no more source cache)
  * should generally be a bit simpler and easier to maintain
  * runs a bit faster :)