Possible memory leak in `convert_result`

BioformatsLoader.jl has had some problems with OOM-errors. I just added a workaround using local frames (ahnlabb/BioformatsLoader.jl@4d4e2d5decd87c8bfd2bfca2fdfbc4214b120977). However, it seems like this could be a bigger issue with the result variable in convert_result, it is unclear to me when that local reference is deleted (it is not made into a JavaObject and does not have the deleteref finalizer) and the memory leak issues in BioformatsLoader.jl are fixed if an explicit JNI.DeleteLocalRef(result) is added as in ahnlabb/JavaCall.jl@94289c5c1cf60e459d0ee4fb5b8c21b6b98282d6 (potentially relevant SO question).

If you think this is indeed the correct approach I can open a PR from ahnlabb/JavaCall.jl@94289c5c1cf60e459d0ee4fb5b8c21b6b98282d6.

As an aside, BioformatsLoader could possibly benefit from increased control over the byte array life cycle and passing raw byte[] references to Java (as is discussed in #83).

Start at convert_result and inspect how the result reference is managed, then compare the proposed JNI.DeleteLocalRef(result) change with the referenced JavaCall.j commit. Reproduce the BioformatsLoader.jl OOM behavior if possible and confirm that the memory leak is resolved without breaking result conversion.