Optimize HistoryTracker using a cached TextDocument.version (#4681)

This replaces `mightChangeDocument`, which sought to do the same sort of optimization in a less effective and far more clumsy way.
Fixes #4665

package.json

@@ -918,11 +918,6 @@
           "type": "boolean",
           "markdownDescription": "When `true` the commands listed below move the cursor to the first non-blank of the line.  When `false` the cursor is kept in the same column (if possible).  This applies to the commands: `<C-d>`, `<C-u>`, `<C-b>`, `<C-f>`, `G`, `H`, `M`, `L`, `gg`, and to the commands `d`, `<<` and `>>` with a linewise operator.",
           "default": true
-        },
-        "vim.experimentalOptimizations": {
-          "type": "boolean",
-          "description": "Increase performance of the extension, especially when navigating or editing large files. This is an experimental option and may cause unpredicted behaviour.",
-          "default": false
         }
       }
     },

src/actions/base.ts

@@ -46,21 +46,6 @@ export class BaseAction {
   private static readonly isSingleNumber: RegExp = /^[0-9]$/;
   private static readonly isSingleAlpha: RegExp = /^[a-zA-Z]$/;
 
-  /**
-   * True if this action has the potential to change the document's text, and therefore trigger an
-   * undo step. This is for optimization purposes - we skip the expensive process of figuring out
-   * what changed to create that undo step if this returns false.
-   *
-   * Most commands do not modify the document, so this is false by default. This means if you add a
-   * command that can modify the document or switch modes, you MUST set mightChangeDocument = true. Otherwise undo
-   * will not work properly.
-   *
-   *
-   * TODO: if all actions were pure, I think this would be unnecessary, as we could deduce it from
-   * vimState.transformations being empty or not.
-   */
-  public mightChangeDocument: boolean = false;
-
   /**
    * Is this action valid in the current Vim state?
    */