Fix LSRA dump of GC Kills

[CarolEidt]

No description provided.

[CarolEidt]

In my PR #679 I attempted to improve the dumping of GC kills (this is where we spill all GC refs across pinvoke calls), but instead broke them. This is actually an improvement over what I'd done in my EHWriteThru branch, and then incorrectly extracted in #679. It will now print something like the following in the LSRA allocation table:

 58.#34      KlGC   Spill ebx  |V8 a|    |    |    |V9 a|    |    |    |    |    |    |
                    Done       |V8 a|    |    |    |V9 a|    |    |    |    |    |    |

If there are no GC refs live in registers, it will print this:

 64.#47      KlGC   None       |    |    |    |    |V9 a|    |    |    |    |    |    |

And if there are multiple it will look something like this:

 886.#716        KlGC   Spill rdi  |     |     |     |V81 a|     |V1  a|     |     |     |V36 i|     |V2  a|V3  i|
                        Spill r14  |     |     |     |V81 a|     |V1  a|     |     |     |V36 i|     |     |V3  i|
                        Done       |     |     |     |V81 a|     |V1  a|     |     |     |V36 i|     |     |V3  i|

Both @briansull and I ran into the unreached() when doing a jitdump. It doesn't make sense for a JITDUMP to ever cause an unreached assert, so I've changed that to just print something to indicate an unknown event was hit.

[CarolEidt]

@dotnet/jit-contrib PTAL

[briansull]

LGTM