feat: upload multiple asset chunks in the same call

[sesi200]

Description

The frontend canister sync now tries to batch multiple small content chunks into a single call using the create_chunks method added in #3898.
This should lead to significantly faster upload times for frontends with many small files.

Also had to bump the api_version of the asset canister to build in backwards compatibility for the asset sync

Fixes SDK-1769

How Has This Been Tested?

Covered by existing e2e.
Manually tested repeatedly with many different average file sizes. Command to test:

for i in $(seq 1 50); do dd if=/dev/urandom of="src/hello_frontend/assets/file_$i.bin" bs=$(shuf -i 1-2000000 -n 1) count=1; done && dfx deploy hello_frontend

Checklist:

• The title of this PR complies with Conventional Commits.
• I have edited the CHANGELOG accordingly.
• I have made corresponding changes to the documentation.

[chenyan-dfinity]

Is icx-asset updated as well? I tested this with one of the caffeine generated directory: https://github.com/chenyan-dfinity/asset-test/tree/main/1723158607637/frontend/dist

$ time ~/src/sdk/target/release/icx-asset upload br5f7-7uaaa-aaaaa-qaaca-cai dist/

On master, it takes 3.475s. On this branch, it takes 3.833s. (Both using wasm from this branch)

My simple chunk upload binary takes 2.158s. I expect icx-asset to be a bit slower because I didn't compute hash and didn't provide gzip files. But at least, it should be faster than master?

[sesi200]

Is icx-asset updated as well?

Yes, it's 'just' a wrapper around ic-asset.

Just tested myself. TLDR: it speeds things up significantly, but only with enough files. Measurements:

Local with dfx start --clean --artificial-delay 500 since I ran out of ICP to test on mainnet. Should be simulating a subnet with no load and 2 blocks per second, which should be close-ish to current ideal mainnet performance.

Project: dfx new, react frontend
icx-asset-pr:       5.38s
icx-asset-master:   5.35s

Project: dfx new, react frontend, plus 500 random files between 0 and 20k bytes, same files for both runs
icx-asset-pr:       8.45s
icx-asset-master:  34.14s

After verifying with debug prints in the asset canister I am pretty certain this happens:

icx-asset-master, hello world:

1. round: create_batch
2. round: 5 parallel calls to create_chunk, all executed in the same round (debug prints show same timestamp)
3. round: commit_batch

icx-asset-pr, hello world:

1. round: create_batch
2. round: create_chunks with 5 chunks in it
3. round: commit_batch

icx-asset-master, hello world + 500 files:

• round 1: create_batch
• then: 500 times create_chunk, up to 25 in parallel. Processes ~20 calls per round
• last 3 rounds: commit_batch since dfx only commits so many changes at a time (execution limits), all wait for the previous call's response

icx-asset-pr, hello world + 500 files:

• round 1: create_batch
• round 2: three calls to create_chunks processed in parallel, all 500 files uploaded
• round 3-5: commit_batch since dfx only commits so many changes at a time, all wait for the previous call's response