Set StreamReadConstraints maxStringLength as early as possible (#355)

Leave the override in Backend as well to avoid duplication. Future
enhancements may enable us to construct a ServiceBackend from argv
alone, allowing this to be reverted.

Add test.

Co-authored-by: Christopher Vittal <[email protected]>

hail/python/test/hail/test_ir.py

@@ -1,4 +1,5 @@
 import functools
+import random
 import re
 import unittest
 from test.hail.helpers import resource, skip_unless_spark_backend, skip_when_service_backend
@@ -774,3 +775,12 @@ def test_locus_interval_encoding():
     _assert_encoding_roundtrip(start)
     _assert_encoding_roundtrip(end)
     _assert_encoding_roundtrip(interval)
+
+
+def test_very_large_ir_deserializes():
+    # see: https://github.com/hail-is/hail/issues/14580
+    #      https://github.com/hail-is/hail/issues/14650
+    large_list = [random.getrandbits(63) for _ in range(5_000_000)]
+    large_lit = hl.literal(large_list, hl.tarray(hl.tint64))
+    round_trip = hl.eval(large_lit)
+    assert large_list == round_trip

hail/src/main/scala/is/hail/backend/service/Main.scala

@@ -1,9 +1,19 @@
 package is.hail.backend.service
 
+import com.fasterxml.jackson.core.StreamReadConstraints
+
 object Main {
   val WORKER = "worker"
   val DRIVER = "driver"
 
+  /* This constraint should be overridden as early as possible. See:
+   * - https://github.com/hail-is/hail/issues/14580
+   * - https://github.com/hail-is/hail/issues/14749
+   * - The note on this setting in is.hail.backend.Backend */
+  StreamReadConstraints.overrideDefaultStreamReadConstraints(
+    StreamReadConstraints.builder().maxStringLength(Integer.MAX_VALUE).build()
+  )
+
   def main(argv: Array[String]): Unit =
     argv(3) match {
       case WORKER => Worker.main(argv)