feat(core): improved task graph concurrency

This commit adds batch-based concurrency to the task graph.

When TaskGraph's process method is called, the requested tasks are
grouped into batches that share one or more keys (including
dependencies). These batches are then queued.

In each iteration of the task graph's main loop, any pending batches
that share no keys with currently in-progress batches are added to the
graph.

This enables e.g. hot reload tasks to be run concurrently with build and
test tasks for their underlying modules, which was one of the primary
motivators behind this change.

Also replaced the concurrencyLimit option for TaskGraph's processTasks
method with an unlimitedConcurrency option. This is useful e.g. when
resolving providers.

Task nodes from batches with unlimited concurrency are processed
regardless of normally available task graph concurrency.

garden-service/src/events.ts

@@ -62,6 +62,10 @@ export type Events = {
   configRemoved: {
     path: string
   }
+  internalError: {
+    timestamp: Date
+    error: Error
+  }
   projectConfigChanged: {}
   moduleConfigChanged: {
     names: string[]

garden-service/src/garden.ts

@@ -506,9 +506,7 @@ export class Garden {
       })
 
       // Process as many providers in parallel as possible
-      const taskResults = await this.processTasks(tasks, {
-        concurrencyLimit: tasks.length,
-      })
+      const taskResults = await this.processTasks(tasks, { unlimitedConcurrency: true })
 
       const failed = Object.values(taskResults).filter((r) => r && r.error)