fix(algorithm): handle multiple prerelease token variants properly

In the case where there are alpha and beta releases, we must only
consider the previous beta release even if alpha releases exist
due to merging into beta release only branches which have no changes
considerable changes from alphas but must be marked otherwise.

Resolves: python-semantic-release#789

semantic_release/version/algorithm.py

@@ -383,6 +383,9 @@ def next_version(
     # bump to produce. However if we're doing a prerelease, we can
     # include prereleases here to potentially consider a smaller portion
     # of history (from a prerelease since the last full release, onwards)
+    # Since there are different types of prereleases, only consider the
+    # prereleases that have the same prerelease token as the current branch's
+    # translator configuration.
 
     # Note that a side-effect of this is, if at some point the configuration
     # for a particular branch pattern changes w.r.t. prerelease=True/False,
@@ -391,7 +394,10 @@ def next_version(
     tag_sha_2_version_lookup = {
         tag.commit.hexsha: (tag, version)
         for tag, version in all_git_tags_as_versions
-        if prerelease or not version.is_prerelease
+        if (
+            (prerelease and version.prerelease_token == translator.prerelease_token)
+            or not version.is_prerelease
+        )
     }
 
     # N.B. these should be sorted so long as we iterate the commits in reverse order