feat: #1994 Add a list of per-request usage data to Usage

[habema]

Resolves #1994

Changes:

• Added RequestUsage class to store per-request details
• Added request_usage_entries field to Usage class
• Modified Usage.add() to automatically create per-request entries for single requests

Sample Usage object after changes:

Usage(
    requests=2,
    input_tokens=158,
    input_tokens_details=InputTokensDetails(cached_tokens=0),
    output_tokens=98,
    output_tokens_details=OutputTokensDetails(reasoning_tokens=0),
    total_tokens=256,
    request_usage_entries=[
        RequestUsage(
            input_tokens=66,
            output_tokens=14,
            total_tokens=80,
            input_tokens_details=InputTokensDetails(cached_tokens=0),
            output_tokens_details=OutputTokensDetails(reasoning_tokens=0),
        ),
        RequestUsage(
            input_tokens=92,
            output_tokens=84,
            total_tokens=176,
            input_tokens_details=InputTokensDetails(cached_tokens=0),
            output_tokens_details=OutputTokensDetails(reasoning_tokens=0),
        ),
    ],
)

[ihower]

@habema 👍 for this small useful feature. I often need the token usage of the last request to estimate how much of the context window is being consumed, rather than looking at the total sum.

By the way, the naming in your PR description doesn’t match the code:

• Class name: PR says IndividualRequestUsage, but the code uses RequestUsage
• Attribute name: PR says individual_requests, but the code uses request_usage_entries

Personally, I feel IndividualRequestUsage and request_usage_entries are a bit verbose. I suggest using RequestUsage (class name) and request_usages (attribute name), clear and simple enough.

[habema]

@ihower thank you. Those mismatches are because of the code review, directly above your comment.

[seratch]

@rm-openai Let me know if you have any comments. I will merge this within a few business days.

[seratch]

@ihower Thanks for sharing your thoughts.

Personally, I feel IndividualRequestUsage and request_usage_entries are a bit verbose. I suggest using RequestUsage (class name) and request_usages (attribute name), clear and simple enough.

request_usages was actually one of the name candidates on my end, but ChatGPT pointed out that it might sound a bit unnatural in English LOL. So, I leaned toward using request_usage_xxx (like entries, list, etc.) instead.