Analyzer version, codegen fix

• Increase minimum version of analyzer.
• Fix generation with the type Never.

Deserialization fix, minor improvements

• Fix generated deserialization code when there is a manually written builder
  with nullable fields.
• Drop dev dependency on quiver.
• Disable all linting of generated code.
• Change generated hash code implementation so it formats better when there are
  many fields.

Bump version of `analyzer`

• Prepare for breaking analyzer changes.
• Bump version of analyzer.

Bump version of `analyzer`

• Bump version of analyzer.

Custom builder improvements; enum improvement

• Fix custom builders in null safe code: allow nested builder fields to be
  nullable.
• Improve custom builders for null safe code: allow abstract setter/getter
  pairs instead of fields. This allows nested builders to have a setter that
  accepts null and a getter that guarantees not to return null, which is
  what auto instantiation of nested builders already provides.
• Allow use of super field initialization in EnumClass.

Fix codegen for generic bounds

• Fix erroneously generated null check for fields with generic bounds.

Null safe `built_value_test`

• Migrate built_value_test to null safety.

Generator fixes, ignore lint

• Fix generation support for optional generic bounds, e.g.
  class Foo<T extends Object?>.
• Fix generation for classes with names starting $.
• Ignore lint unnecessary_lambdas in generated code.

Generator improvements

• Change generated build methods to return only public types, creating
  _build methods that return the generated impl types. This means dartdoc
  will no longer reference the generated types.
• Ignore the no_leading_underscores_for_local_identifiers lint in generated
  code.
• Migrated built_value_generator to null safety. This is purely an internal
  change, the generator can still generate legacy code as and when needed.

Fix serializer generation corner case

• Bug fix: fix a corner case with generics that had incorrect serializer generation.