fix: Use lazy Stream operations in decode_json_stream to avoid memory spike (#3799)

## Summary

Changed `Enum.filter` and `Enum.map` to `Stream.filter` and `Stream.map`
in `decode_json_stream/1` to prevent loading the entire shape log into
memory at once.

## Problem

The `decode_json_stream/1` function in `Materializer` was breaking lazy
evaluation:

```elixir
stream
|> Stream.map(&Jason.decode!/1)      # lazy ✓
|> Enum.filter(...)                  # EAGER - materializes entire stream!
|> Enum.map(...)                     # eager
```

When a shape has a long history (millions of operations), this causes a
massive memory spike during Materializer startup because `Enum.filter`
forces the entire stream into memory before processing continues.

## Solution

Replace the eager `Enum` functions with lazy `Stream` equivalents:

```elixir
stream
|> Stream.map(&Jason.decode!/1)      # lazy ✓
|> Stream.filter(...)                # lazy ✓
|> Stream.map(...)                   # lazy ✓
```

The downstream `apply_changes/2` function uses `Enum.reduce`, which
correctly consumes the stream one element at a time, so records now flow
through the entire pipeline lazily.

## Test Plan

- [x] All existing tests pass

---
Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>

Author: robacourt

.changeset/lazy-stream-decode.md

@@ -0,0 +1,5 @@
+---
+"@core/sync-service": patch
+---
+
+Fix memory spike during Materializer startup by using lazy stream operations instead of eager Enum functions in `decode_json_stream/1`.

packages/sync-service/lib/electric/shapes/consumer/materializer.ex

@@ -222,10 +222,10 @@ defmodule Electric.Shapes.Consumer.Materializer do
   defp decode_json_stream(stream) do
     stream
     |> Stream.map(&Jason.decode!/1)
-    |> Enum.filter(fn decoded ->
+    |> Stream.filter(fn decoded ->
       Map.has_key?(decoded, "key") || Map.has_key?(decoded["headers"], "event")
     end)
-    |> Enum.map(fn
+    |> Stream.map(fn
       %{
         "key" => key,
         "value" => value,