[feature] Conversations API (#3013)

* Implement conversations API

* Sort and page conversations by last status ID

* Appease linter

* Fix deleting conversations and statuses

* Refactor to make migrations automatic

* Lint

* Update tests post-merge

* Fixes from live-fire testing

* Linter caught a format problem

* Refactor tests, fix cache

* Negative test for non-DMs

* Run conversations advanced migration on testrig startup as well as regular server startup

* Document (lack of) side effects of API method for deleting a conversation

* Make not-found check less nested for readability

* Rename PutConversation to UpsertConversation

* Use util.Ptr instead of IIFE

* Reduce cache used by conversations

* Remove unnecessary TableExpr/ColumnExpr

* Use struct tags for both unique constraints on Conversation

* Make it clear how paging with GetDirectStatusIDsBatch should be used

* Let conversation paging skip conversations it can't render

* Use Bun NewDropTable

* Convert delete raw query to Bun

* Convert update raw query to Bun

* Convert latestConversationStatusesTempTable raw query partially to Bun

* Convert conversationStatusesTempTable raw query partially to Bun

* Rename field used to store result of MaxDirectStatusID

* Move advanced migrations to their own tiny processor

* Catch up util function name with main

* Remove json.… wrappers

* Remove redundant check

* Combine error checks

* Replace map with slice of structs

* Address processor/type converter comments

- Add context info for errors
- Extract some common processor code into shared methods
- Move conversation eligibility check ahead of populating conversation

* Add error context when dropping temp tables

1 files changed, 28 insertions, 0 deletions

diff --git a/internal/cache/wrappers.go b/internal/cache/wrappers.go
index edeea9bcd..9cb4fca98 100644
--- a/internal/cache/wrappers.go
+++ b/internal/cache/wrappers.go
@@ -158,6 +158,34 @@ func (c *StructCache[T]) LoadIDs(index string, ids []string, load func([]string)
 	})
 }
 
+// LoadIDs2Part works as LoadIDs, except using a two-part key,
+// where the first part is an ID shared by all the objects,
+// and the second part is a list of per-object IDs.
+func (c *StructCache[T]) LoadIDs2Part(index string, id1 string, id2s []string, load func(string, []string) ([]T, error)) ([]T, error) {
+	i := c.index[index]
+	if i == nil {
+		// we only perform this check here as
+		// we're going to use the index before
+		// passing it to cache in main .Load().
+		panic("missing index for cache type")
+	}
+
+	// Generate cache keys for two-part IDs.
+	keys := make([]structr.Key, len(id2s))
+	for x, id2 := range id2s {
+		keys[x] = i.Key(id1, id2)
+	}
+
+	// Pass loader callback with wrapper onto main cache load function.
+	return c.cache.Load(i, keys, func(uncached []structr.Key) ([]T, error) {
+		uncachedIDs := make([]string, len(uncached))
+		for i := range uncached {
+			uncachedIDs[i] = uncached[i].Values()[1].(string)
+		}
+		return load(id1, uncachedIDs)
+	})
+}
+
 // Store: see structr.Cache{}.Store().
 func (c *StructCache[T]) Store(value T, store func() error) error {
 	return c.cache.Store(value, store)