feat: add per-component probe overrides

[snigdhasambitak]

What

Per-component startupProbe / readinessProbe / livenessProbe override blocks on manager, scheduler, seedClient, client, injector. Component fields win; unset fields fall back to the existing global.

Why

#501 had per-component startupProbe. #498's squash-merge collapsed it into one global block, not called out in any commit message. One timing fits no component well: seedClient cold-boots in 60-120s and needs slack, manager wants tight liveness. We hit manager liveness flap on cold boot in our clusters today.

Example

# enable startup on slow components only
client:
  startupProbe:
    enable: true
seedClient:
  startupProbe:
    enable: true
    failureThreshold: 90  # ~7.5 min for cold cache hydration
# manager keeps the global default (startup disabled, tight liveness)

No override = zero diff vs main (verified by helm template byte-diff). Helper uses hasKey since sprig merge drops explicit zero values. Chart 1.6.26 → 1.6.27, README regenerated.

[snigdhasambitak]

@imeoer @gaius-qi @CormickKneey could one of you add the enhancement label so the Classify PR check passes? External contributors can't apply labels themselves. Thanks.

[snigdhasambitak]

Any reviews or feedbacks? This will help us to override the probes. So can this be reviewed?