Scaling behavior (EP plans) - low frequency input events, many (same) ActivityFunctions launched

Hi,

First, many thanks for Azure functions the ecosystem!

After several days of tests and research, we still struggle to adjust the scaling behavior to fit our needs.

- We run a ML inference algo in a "Python + durable functions + custom docker + ElasticPremium" context.
- We have a low frequency event input. Basically users upload a CSV file, and we get it using a BlobTrigger (we plan to use HTTPTriggers later).
This file can be quite long (up to 1000 lines), and each line is a computation.
- For each line, we launch a computation (In an activity function), and this computation is quite CPU intensive (>70% CPU) and takes 2 to 4 mins to complete (on EP4 plan)

Sometimes, we're OK with a computation 'batch' running several hours on 1 instance, but for some business needs, we would like to accelerate and benefit from the scaling behavior.

Last time, we had 1 'batch' of 250 computations to do quickly (1 csv file with 250 lines) :
- we maxed the plan to EP4 (4 cores)
- we set up the plan scale to "Always ready": 1, "Maximum burst" & "Maximum scale out limit" to 10 (see screenshot)
- we set up `"maxConcurrentActivityFunctions":1` in `host.json`, to stimulate instances spawning
- the orchestrator Function is in fan-out / fan-in pattern (`results =yield context.task_all(tasks)`)

**But there was only two instances that had spawned during the 3-4 hours run (we expected more).**

**We've read the documentation several times (and lot of GH issues) about scaling, but we're uncertain on what to try next to achieve the desired scaling behavior.**