Tilt adds Dockerfile to image if user specifies `COPY .`

A user had an issue copying . (in their container) into a volume: they got permission errors b/c their Dockerfile was surprise in that director, with permissions 0000.

The Dockerfile-without-permissions ended up in the container b/c of a quirk of Tilt: we add the Dockerfile to the context tar archive b/c the Docker build protocol requires that the Dockerfile exist somewhere in the archive. However, Tilt's solution is inelegant: we add the Dockerfile to the root of the context, so if the user's Dockerfile calls COPY . ..., it picks up that Dockerfile and adds it to the image, sometimes unexpectedly. What's more, we add the Dockerfile w/o specifying permissions, so the permissions zero out and we're left with a file with no permissions, causing the issue the user ran into.

#3594 fixes the latter issue -- we now tar in the Dockerfile with permissions 0644 so users can at least interact with it. The ideal solution here would be to replicate Docker's behavior of putting the Dockerfile into the context with a random name and then .dockerignoreing that name so that, while the DF is in the archive (as is needed for building the image), it doesn't get copied into the container if the user runs COPY . ....

Start in internal/build/tar.go around line 251, then compare the referenced Docker CLI context handling. Trace how Tilt adds the Dockerfile to the context archive and how Dockerfile and .dockerignore entries are represented. Done means a Docker build still receives the Dockerfile, while COPY . does not include that injected file.