Fix/forced variations double format exception

[vazarkevych]

fix: handle int, double, and string values in forcedVariationsMap to prevent FormatException

Problem

forcedVariationsMap is typed as Map<String, dynamic>, so when variation values are deserialized from JSON, numeric values can arrive as double (e.g. 1.0) instead of int. The previous implementation converted any value to a string and called int.parse(), which throws a FormatException on inputs like "1.0":

int.parse(1.0.toString()); // FormatException: Invalid radix-10 number

This was a runtime crash for any integration passing numeric variation indices through JSON.

---

Solution

Replace the int.parse(.toString()) call in experiment_evaluator.dart with explicit type-aware parsing:

• int → used directly
• double → converted via .toInt()
• String → tried with int.tryParse, then double.tryParse(...).toInt() as fallback
• Unsupported types (e.g. bool, objects) → log a warning and skip the forced variation gracefully instead of crashing

---

Test Cases Added

Input	Result
1 (int)	variationID = 1
1.0 (double)	variationID = 1
"1" (string)	variationID = 1
"1.0" (string)	variationID = 1
"abc"	no crash, forced variation skipped
true	no crash, forced variation skipped

---

Backward Compatibility

Existing integrations passing integer values are unaffected — the int branch is the first check and returns immediately.

[madhuchavva]

Inline review comment on forced variation parsing semantics.

[madhuchavva]

Good catch on int.parse(value.toString()); that crash is worth fixing. I think this parsing should be stricter before merge: double.toInt() silently turns values like 1.9 into variation 1, and string parsing via double.tryParse(rawValue)?.toInt() has the same issue. Forced variation IDs should be exact finite integer indexes, so we should only accept integer-valued numbers/strings (1, 1.0, "1", "1.0") and skip/log fractional or non-finite values. It would also be useful to validate the parsed index against experiment.variations before returning the forced result.

[vazarkevych]

Done — updated the parsing to only accept integer-valued numbers for doubles, same for string-parsed doubles and added bounds validation against experiment.variations.length before applying the forced result. Fractional or out-of-bounds values are now logged and skipped. Tests added for all new cases.

[madhuchavva]

Requesting changes based on the inline review comments. The notes call out functional gaps or compatibility/documentation risks that should be addressed before merge.