[#1049] Fix gtm_jlong_t return type handling on 32-bit armv6l (regres…

…sion in 9f916f6)

Background
----------
* The `java/callouts` subtest failed on armv6l with the following diff.

  ```diff
  7a8,67
  > 16c16
  > < Job returned -336177440569076945
  > ---
  > > Job returned -1277316960
  .
  .
  ```

Issue
-----
* Before 9f916f6, on 32-bit armv6l, a `ydb_jlong` type return value was handled in `sr_unix/op_fnfgncal.c`
  by converting the 8-byte contents of the pointer to the return value. This was done whether `starred`
  was TRUE or not. That is, the handling was the same for `gtm_jlong_t` and `gtm_jlong_t *`.

* But after 9f916f6, on 32-bit armv6l, a `ydb_jlong` type return value was handled differently for
  `gtm_jlong_t` and `gtm_jlong_t *`. Just like it was done on 64-bit systems. While the different
  handling is correct and needed for 64-bit systems, it is incorrect for 32-bit systems.

Fix
---
* The fix is to restore the same handling for `gtm_jlong_t` and `gtm_jlong_t *` for 32-bit systems.
  That is, there should be no `if (starred)` logic in the `#ifndef GTM64` case for `case ydb_jlong:`
  in `sr_unix/op_fnfgncal.c` like is there for the `#ifdef GTM64` case..

sr_unix/op_fnfgncal.c

@@ -353,10 +353,18 @@ STATICFNDEF void extarg2mval(void *src, enum ydb_types typ, mval *dst, boolean_t
 				MV_FORCE_MVAL(dst, s_int_num);
 				break;
 			case ydb_jlong:
+				/* On 64-bit systems, "gtm_jlong_t *" is passed in using a pointer to the actual 8-byte value
+				 * whereas a "gtm_jlong_t" is passed in as is. On 32-bit systems though, both "gtm_jlong_t"
+				 * and "gtm_jlong_t *" are passed in using a pointer to the actual 8-byte value.
+				 */
+#				ifdef GTM64
 				if (starred)
-					s_int64_num = *((ydb_int64_t*)src);
+					s_int64_num = *((ydb_int64_t *)src);
 				else
-					s_int64_num = (ydb_int64_t)(intszofptr_t)src;
+					s_int64_num = (ydb_int64_t)src;
+#				else
+				s_int64_num = *((ydb_int64_t *)src);
+#				endif
 				MV_FORCE_64MVAL(dst, s_int64_num);
 				break;
 			case ydb_jfloat: