Enable low-latency streaming for fresh installs

[a692570]

Summary

Enable Deepgram streaming by default when Bolo uses its default deepgram/nova-3 STT model. Keep explicit off, AssemblyAI selection, and custom-model behavior intact.

Why this change

Fresh installs used batch transcription while the maintainer configuration enabled streaming, so new users did not receive the same low-latency path.

Testing

• Tested locally on macOS
• cargo fmt --all -- --check
• cargo test --release (53 passed)
• python3 -m pytest -q (43 passed)
• Shell scripts pass bash -n
• Reviewed the configuration fallback behavior

Documentation

• README updated for the new default and explicit opt-out
• No UI change

Notes

Selecting a non-default BOLO_STT_MODEL leaves streaming disabled unless the user explicitly selects a streaming provider.

[entelligence-ai-pr-reviews]

EntelligenceAI PR Summary

This PR makes Deepgram streaming (deepgram/nova-3) the default STT mode, replacing the previous opt-in behavior, and refactors the related configuration logic for testability.

• README.md: Updated docs so BOLO_STT_STREAMING='off' is now required to disable streaming; added clarification that non-default BOLO_STT_MODEL keeps streaming off unless overridden
• src/main.rs: load_streaming_provider now accepts stt_model; new streaming_provider_from_config function implements smart defaulting (Deepgram streaming auto-enabled for deepgram/nova-3, explicit config values still take precedence)
• src/main.rs: Unit tests added to cover all branching scenarios of the new default behavior

---

Confidence Score: 5/5 - Safe to Merge

Safe to merge — this PR cleanly promotes Deepgram streaming (deepgram/nova-3) to the default STT mode and refactors load_streaming_provider into a testable streaming_provider_from_config function with no identified logic, security, or runtime issues. The behavioral change (opt-out via BOLO_STT_STREAMING='off' rather than opt-in) is well-documented in README.md, including the edge case where a non-default BOLO_STT_MODEL suppresses streaming unless explicitly overridden. No review comments were generated and heuristic analysis found zero issues at any severity level.

Key Findings:

• streaming_provider_from_config in src/main.rs correctly accepts stt_model as a parameter, enabling isolated unit testing of the streaming decision logic without environment side effects — a meaningful improvement in testability.
• The README update accurately captures the new default-on semantics and the interaction between BOLO_STT_STREAMING and BOLO_STT_MODEL, reducing the risk of operator misconfiguration on fresh installs.
• No critical, significant, or medium-severity issues were surfaced by either automated heuristics or manual review, and there are no pre-existing unresolved comments carried into this PR.

Files requiring special attention

• src/main.rs
• README.md

[entelligence-ai-pr-reviews]

Walkthrough

This PR changes the default behavior of STT streaming so that Deepgram (deepgram/nova-3) streaming is enabled by default instead of requiring explicit opt-in. Documentation is updated to reflect the new default, and the streaming provider loading logic is refactored into a testable function with smart defaulting. Unit tests covering all branching scenarios are added.

Changes

File(s)	Summary
README.md	Updated STT streaming documentation to reflect Deepgram streaming is now on by default; revised instructions for disabling streaming via BOLO_STT_STREAMING='off'; added note about non-default BOLO_STT_MODEL keeping streaming off unless explicitly set.
src/main.rs	Refactored load_streaming_provider to accept stt_model parameter; extracted streaming_provider_from_config with smart defaulting logic (auto-enables Deepgram streaming for deepgram/nova-3); added unit tests for all branching scenarios.

Sequence Diagram

This diagram shows the interactions between components:

sequenceDiagram
    title STT Streaming Provider Configuration Flow

    participant Config as Config::new()
    participant LSP as load_streaming_provider()
    participant SPFC as streaming_provider_from_config()
    participant Env as Environment Variables

    Config->>Env: load_env_value("BOLO_STT_MODEL")
    Env-->>Config: stt_model (or "deepgram/nova-3")

    Config->>LSP: load_streaming_provider(&stt_model)
    activate LSP

    LSP->>Env: load_env_value("BOLO_STT_STREAMING")
    Env-->>LSP: value (or None)

    opt value is None
        LSP->>Env: load_env_value("BOLO_STREAMING_STT")
        Env-->>LSP: value (or None)
    end

    LSP->>SPFC: streaming_provider_from_config(value, stt_model)
    activate SPFC

    alt value is None
        Note over SPFC: Determine default from stt_model
        alt stt_model == "deepgram/nova-3"
            Note over SPFC: default = "deepgram"
        else stt_model is anything else
            Note over SPFC: default = "off"
        end
    end

    alt resolved value == "deepgram" / "nova-3" / "nova3"
        SPFC-->>LSP: Some(StreamingProvider::Deepgram)
    else resolved value == "assemblyai" / "assembly" / "on"
        SPFC-->>LSP: Some(StreamingProvider::AssemblyAi)
    else resolved value == "off" or unrecognized
        SPFC-->>LSP: None
    end

    deactivate SPFC
    LSP-->>Config: Option<StreamingProvider>
    deactivate LSP

    Note over Config: Store stt_model and streaming_stt in Config struct

Loading