[BugFix][CI] make build_bisect_extra_args always return success (#15096)

The function's last `[ -n ... ] && ...` short-circuits to a non-zero
status when its env var is unset (the common case --
BISECT_CONFIG_BASE_PATH defaults to '', and the /nightly command does
not expose --config-base-path to override it). That return value becomes
the function's exit status, and since the callers run under `set -euo
pipefail`, the script aborts before `python -m tools.bisect.auto_bisect`
is ever launched.

Two manifestations seen in nightly:
- single-node (aop_process.sh): the bisect never starts -- exit 1 at the
build_bisect_extra_args call, before the "=== Running auto bisect ==="
line is printed.
- multi-node (run.sh worker branch): the worker is killed at the same
call under set -e, so the bisect worker agent never starts, ready_1.json
is never written, the leader's 60min barrier times out, the bad commit
is skipped as an environment error, and the bisect aborts (exit 2).

The helper only appends args to BISECT_EXTRA_ARGS and has no failure
mode, so always returning 0 is the correct fix. (The leader's
aop_pipeline in run.sh happened to be safe because it is called under
`set +e`, but the worker branch and aop_process.sh are not.)

### What this PR does / why we need it?

### Does this PR introduce _any_ user-facing change?

### How was this patch tested?

- vLLM main:
https://github.com/vllm-project/vllm/commit/ba07e4a48fc951300d97eb506217dd530583dea3

Signed-off-by: chenzeyu <2978509328@qq.com>