filter/addPoissonNoise hangs up with large pixel values

Hello,

I have been working on a project for the NIH that uses [pyimagej](https://github.com/imagej/pyimagej) to run ImageJ ops from python. Yesterday I was working on developing a plugin which calls the [filter-addpoissonnoise op](https://github.com/imagej/imagej-ops/blob/master/src/main/java/net/imagej/ops/filter/addPoissonNoise/AddPoissonNoiseRealType.java). During testing I determined that the op would hang up seemingly doing nothing when the input images had large pixel values. I believe this problem can be summarized as a precision/underflow problem when very large lambda values (expected number of successes in a given interval) are introduced to the Knuth algorithm. This has been documented before in wiki's [Generating Poisson-distributed random variables section](https://en.wikipedia.org/wiki/Poisson_distribution). As lambda becomes large exp(-lambda) approaches 0 and becomes so small it is evaluated as exactly 0. This value is checked in a while loop against another value which is never less than 0: `while p >= L: do stuff` where `L` = 0 and `p` is never less than 0. Therefore the loop runs forever.

Start with src/main/java/net/imagej/ops/filter/addPoissonNoise/AddPoissonNoiseRealType.java, especially the loop implementing the Knuth algorithm. Reproduce the hang with an image containing large pixel values and determine how the underflowed exp(-lambda) affects termination. Done means the op terminates and produces valid Poisson noise for large lambda values.

Written by the indexing model from the issue text.