`SCHED_DEADLINE` range check allows value equal to 2^63 (off-by-one vs. cited spec)

I asked Claude Opus 4.8

```
Is there any mistake bug in src/libcrun/scheduler.c ?
I wonder if there is any mismatch between the if statement checks and the error messages.
```

Claude wrote a lot of text. I then wrote:

```
Can you draft a minimal issue (in markdown format) for the issue you found?
```

Click me for claude output

## `SCHED_DEADLINE` range check allows value equal to 2^63 (off-by-one vs. cited spec)

### Description
In `src/libcrun/scheduler.c`, `diagnose_scheduler_failure()` validates the
`SCHED_DEADLINE` runtime/deadline/period against a `[min, max]` range. The code
comment quotes sched(7) as requiring values "at least 1024 ... and less than
2^63", i.e. the valid range is `[1024, 2^63)`. However the upper-bound guard uses
`> max` with `max = 1ULL << 63`, so a value of exactly `2^63` passes the check
even though the spec says it must be *less than* `2^63`. The error message
("must be between 1024 and 2^63") is worded consistently with the buggy check
rather than the spec.

### Location
`src/libcrun/scheduler.c`, in `diagnose_scheduler_failure()` (runtime, deadline,
and period range checks).

### Current code
```c
const uint64_t min = 1024;
const uint64_t max = 1ULL << 63; /* == 2^63 */

if (attr->sched_runtime < min || attr->sched_runtime > max)
return crun_make_error (err, errno,
"sched_setattr: `SCHED_DEADLINE` runtime (%" PRIu64
") must be between %" PRIu64 " and %" PRIu64,
attr->sched_runtime, min, max);
```

### Expected vs. actual
- Expected: a value equal to `2^63` is rejected (spec: `< 2^63`).
- Actual: `> max` lets `2^63` through; only `> 2^63` is rejected.

### Suggested fix
Reject the endpoint and align the message, e.g.:
```c
if (attr->sched_runtime < min || attr->sched_runtime >= max)
return crun_make_error (err, errno,
"sched_setattr: `SCHED_DEADLINE` runtime (%" PRIu64
") must be >= %" PRIu64 " and < %" PRIu64,
attr->sched_runtime, min, max);
```
(or set `max = (1ULL << 63) - 1` and keep `>`). The same change applies to all
three range checks, which share `min`/`max`.

### Notes
Low severity: this affects diagnostic accuracy rather than behavior — the kernel
would reject such a value anyway, and crun would fall through to the generic
"invalid parameters" error.

Start in src/libcrun/scheduler.c at diagnose_scheduler_failure() and inspect the shared SCHED_DEADLINE runtime, deadline, and period range checks against the cited sched(7) bounds. Confirm that the exact 2^63 endpoint is rejected for all three values and that the resulting error messages describe the inclusive lower and exclusive upper bounds.

Written by the indexing model from the issue text.