Tweak condition for TEXTDECODER=2 default (#24349)

- When using pthreads, our `TextDecoder` usage already correctly handles
shared memory through the `getUnsharedTextDecoderView` helper, so the
check and the comment are outdated.
- When targeting Audio Worklets, we already have assertion that
TEXTDECODER!=2 as it's not compatible with that environment, so opt-out
from the size optimisation there instead.

Author: RReverser

test/other/codesize/test_codesize_minimal_pthreads.gzsize

@@ -1 +1 @@
-3988
+3795

test/other/codesize/test_codesize_minimal_pthreads.jssize

@@ -1 +1 @@
-8264
+7883

tools/link.py

@@ -1008,11 +1008,12 @@ def limit_incoming_module_api():
 
   # Default to TEXTDECODER=2 (always use TextDecoder to decode UTF-8 strings)
   # in -Oz builds, since custom decoder for UTF-8 takes up space.
-  # In pthreads enabled builds, TEXTDECODER==2 may not work, see
-  # https://github.com/whatwg/encoding/issues/172
   # When supporting shell environments, do not do this as TextDecoder is not
   # widely supported there.
-  if settings.SHRINK_LEVEL >= 2 and not settings.SHARED_MEMORY and \
+  # In Audio Worklets TextDecoder API is intentionally not exposed
+  # (https://github.com/WebAudio/web-audio-api/issues/2499) so we also need to
+  # keep the JavaScript-based fallback.
+  if settings.SHRINK_LEVEL >= 2 and not settings.AUDIO_WORKLET and \
      not settings.ENVIRONMENT_MAY_BE_SHELL:
     default_setting('TEXTDECODER', 2)