fix(stream): gate no-size read() head-chunk behavior on flowing mode

[proggeramlug]

Summary

Unbreaks cargo-test on main (every open PR is currently red on it).

PR #5017 changed Readable.read() with no size to return only the head chunk of the internal buffer unconditionally. Node's howMuchToRead(NaN) only does that when the stream is flowing:

if (NumberIsNaN(n)) {
  if (state.flowing && state.length) return state.buffer.first().length;
  return state.length; // paused: drain the whole buffer
}

A paused stream drains the entire buffer and returns it as a single value, which is exactly what the two now-failing unit tests assert:

• node_stream::tests_extra::readable_unshift_prepends_chunk_and_returns_hwm_signal
• node_stream::tests_extra::readable_unshift_after_eof_before_end_prepends_chunk

(push('world'); unshift('hello '); read() → 'hello world', verified against real Node.)

Fix

read_stream_available_default keeps the #5017 head-chunk path when readable_is_flowing(stream) and restores the pre-#5017 full-drain concat path (drain_whole_buffer, including the string-encoding branch) when paused. Sized read(n) is untouched.

Validation

• cargo test --release -p perry-runtime --lib node_stream → 68 passed, 0 failed (both previously-failing tests pass).
• Full perry-runtime --lib suite: only the two known macOS-local pre-existing failures remain (date full_year, Date.prototype.toJSON); both also fail on unmodified origin/main locally and are green in Linux CI.

Note: the remaining main reds are tracked separately — #5029 (gc_write_barrier_stress) and #5030 (h1_buffer_alias_negative); fixes in flight.

No version bump / changelog per external-PR convention — maintainer folds metadata at merge.

[proggeramlug]

Superseded by #5033 (same commit, stacked with the #5030 fix and the #5029 test gating so all required checks can pass together).