feat: Cache misses are reported with a reason

When `skaffold build` or `skaffold dev` decides to rebuild an artifact, the only
output is:

```
Checking cache...
- app: Not found. Building
```

That line is the same whether:

- a source file the artifact depends on changed,
- a build arg or the artifact's config changed,
- or nothing changed at all and the previously built image was simply pruned
from the local daemon or evicted from the registry.

The last case is the confusing one: nothing the user did caused it, but the
output is identical to a real source change. Diagnosing an unexpected rebuild
today means building a patched skaffold and printing the hash inputs by hand.

Two things would help, and I have both working locally:

1. Report the reason on the existing line. The lookup already knows whether the
artifact's current hash had an entry in the cache, which separates "the
inputs produced a hash we have never built" from "we built this, but the
image is gone":

```
- app: Not found. Building (no cached build for the current inputs)
- api: Not found. Building (cached image is no longer available)
```

No new flags, files or I/O — it is derived from state the lookup already has.

2. Behind an opt-in flag, record the inputs that feed each artifact's hash and
diff them against the previous run, so the miss names what changed:

```
- app: Not found. Building (inputs changed)
~ src/handler.go
+ src/middleware.go
- src/old_handler.go
```

Reproduce the cache miss with `skaffold build` or `skaffold dev`, then start at the cache lookup that already distinguishes the current hash from cache availability. Done means the existing build message explains those two miss reasons, and an opt-in mode reports changed hash inputs between runs.

Written by the indexing model from the issue text.