* Fix Creation policy and Idle duration

* Add fleet configurations for multi-node examples

* Add Create fleet to quickstart

* Fix Idle duration

* Move Fleets up

* Do not mention autocreated fleets

* Fix pronoun

* Update docs/docs/quickstart.md

Co-authored-by: Andrey Cheptsov <[email protected]>

* Update docs/docs/quickstart.md

Co-authored-by: Andrey Cheptsov <[email protected]>

* Update docs/docs/quickstart.md

Co-authored-by: Andrey Cheptsov <[email protected]>

* Make Fleet section collapsed by default

* [Docs] Improve fleet documentation to reflect fleet-first UX changes

---------

Co-authored-by: Andrey Cheptsov <[email protected]>
Co-authored-by: peterschmidt85 <[email protected]>

* [CLI] Add color to the STATUS column

* [CLI] Show `RESOURCES` only in verbose mode, otherwise only show `GPU`. Also move `(spot)` to `PRICE.

* [CLI] Fix tests

* [CLI] Show `replica=...` and `job=...` only when needed.

* [CLI] Show `replica=...` and `job=...` only when needed. (bugfix)

* [CLI] Show `replica=...` and `job=...` only when needed. (cleanup+bugfix)

The patch was originally introduced to address a
registry authentication issue, which has since
been fixed upstream in version 12.1.1.

Previously, KubernetesCompute only used GPU from the first offer to
set node affinity, and if that type of GPU was not available (e.g.,
another job or even some non-dstack pod had already taken it), the
job eventually failed with FAILED_TO_START_DUE_TO_NO_CAPACITY, even
if there were other GPUs matching the run spec requirements.

Now, we inspect all nodes to request all suitable GPUs (any of).

In addition, we now use upper bounds of Ranges (CPU, memory, disk)
as limits except for GPU, which cannot have request =/= limit (as it
cannot be overcommited).

Part-of: #3126

…ner consume all ram and hang (#3263)

Co-authored-by: Dmitry Meyer <[email protected]>

Previously, `dstack` did not return any non-GPU
offers if GPU name, memory, or vendor requirements
are set, regardless of the GPU count requirement.

```shell
> dstack offer --gpu 0..:nvidia
 #   BACKEND             RESOURCES                                   INSTANCE TYPE  PRICE
 1   aws (eu-west-2)     cpu=4 mem=16GB disk=100GB T4:16GB:1 (spot)  g4dn.xlarge    $0.1879
 2   aws (us-east-1)     cpu=4 mem=16GB disk=100GB T4:16GB:1 (spot)  g4dn.xlarge    $0.1914
 3   aws (ca-central-1)  cpu=4 mem=16GB disk=100GB T4:16GB:1 (spot)  g4dn.xlarge    $0.1931
 <...>
```

Now, it Includes non-GPU offers if the lower GPU
count bound is set to `0`.

```shell
> dstack offer --gpu 0..:nvidia
 #     BACKEND             RESOURCES                                         INSTANCE TYPE  PRICE
 1     aws (eu-north-1)    cpu=2 mem=8GB disk=100GB (spot)                   t3.large       $0.0281
 2     aws (us-west-1)     cpu=2 mem=8GB disk=100GB (spot)                   t3.large       $0.0297
 3     aws (eu-west-2)     cpu=2 mem=8GB disk=100GB (spot)                   t3.large       $0.0298
 <...>
 1752  aws (ca-central-1)  cpu=192 mem=2048GB disk=100GB H100:80GB:8 (spot)  p5.48xlarge    $63.296
 1753  aws (us-west-1)     cpu=192 mem=2048GB disk=100GB H100:80GB:8         p5.48xlarge    $68.8
 1754  aws (eu-west-2)     cpu=192 mem=2048GB disk=100GB H100:80GB:8         p5.48xlarge    $71.552
```