docs: changelog + doc updates for unreleased PRs

[jrob5756]

Summary

Updates the [Unreleased] changelog entry to cover all 18 PRs merged since v0.1.10, and brings the user-facing docs in line with new public surface introduced by those PRs.

Changelog

Adds the 9 PRs that landed since the last changelog refresh:

• fix(engine): ensure workflow.input available for script/sub-workflow templates in explicit mode #119 — workflow.input available to script/workflow templates in explicit mode
• feat(engine): add workflow.dir, workflow.file, workflow.name template variables #121 — workflow.dir / workflow.file / workflow.name template variables
• feat(engine): auto-parse script agent JSON stdout into output fields #122 — script agent JSON stdout auto-parsing
• fix(engine): use type-appropriate zero values for optional input defaults #123 — type-appropriate zero values for optional input defaults
• feat(validate): catch template reference errors before runtime #125 — conductor validate semantic checks for template references
• feat: breadcrumb navigation, subworkflow isolation, and stop reliability #113 + fix(web): dedupe phantom edges, for-each dive-in nav, PR #113 review nits (#145) #146 — dashboard breadcrumbs, subworkflow isolation, stop reliability (and follow-up nits)
• feat(dialog): agent dialog mode with web dashboard support #130 — agent dialog mode
• feat(gates): markdown rendering and auto-linkification in human gate prompts #131 — Markdown rendering and auto-linkification in human gate prompts
• feat: add workspace instructions support (--workspace-instructions) #141 — --workspace-instructions and --instructions

Doc updates

docs/workflow-syntax.md

• Add metadata and instructions to the workflow configuration block (feat(config): add optional metadata dict to workflow definition #107, feat: add workspace instructions support (--workspace-instructions) #141).
• Sub-Workflow Steps: add input_mapping and max_depth; fix stale claims that circular references are rejected (now allowed via depth limit) and that workflow steps cannot be used in for_each groups (now allowed). Add a for_each-with-sub-workflow example. (feat(composition): dynamic sub-workflow inputs via input_mapping (#101) #109, feat(composition): allow sub-workflows in for_each groups (#102) #110, feat(composition): allow self-referential sub-workflows with depth tracking (#103) #111)
• Script Steps: document JSON stdout auto-parsing into output fields with a route-on-parsed-field example. (feat(engine): auto-parse script agent JSON stdout into output fields #122)
• Workflow Inputs: document type-appropriate zero values for optional inputs without an explicit default. (fix(engine): use type-appropriate zero values for optional input defaults #123)
• New 'Workflow Metadata Variables' subsection covering workflow.dir / workflow.file / workflow.name. (feat(engine): add workflow.dir, workflow.file, workflow.name template variables #121)
• Update on_start hook context list to include the new vars.

docs/cli-reference.md

• Document --metadata/-m, --workspace-instructions, and --instructions flags on conductor run. (feat(config): add optional metadata dict to workflow definition #107, feat: add workspace instructions support (--workspace-instructions) #141)
• Add 'Metadata and Instructions' examples block.
• Expand 'Validation Checks' to describe template-reference errors and explicit-mode warnings. (feat(validate): catch template reference errors before runtime #125)

docs/providers/claude.md, docs/providers/comparison.md

• Replace stale 'All models support a 200K token context window' / '200K (all models)' claims with notes that the dashboard now sources context_window_max from each provider's SDK at runtime. (feat(engine): source context_window_max from provider SDKs at runtime #144)

README.md

• Refresh Features list (sub-workflow composition, dialog mode, workspace instructions, breadcrumb navigation, enhanced validate).
• Add --metadata, --workspace-instructions, --instructions to the run options table.

What I did NOT touch

• PRs fix(copilot): pass streaming=True to SDK to prevent tool-call truncation #129, fix(copilot): infer nested prompt schema from output definitions #100, fix(engine): coerce Python literal "True"/"False"/"None" in workflow output #139, fix(pricing): prevent cross-family fuzzy match + warn on fallback #143, fix(install): run uv tool update-shell so conductor is on PATH for new shells #142 are pure bug fixes / internal — already covered in the prior changelog and need no doc updates.
• PRs feat(dialog): agent dialog mode with web dashboard support #130 (dialog) and feat(gates): markdown rendering and auto-linkification in human gate prompts #131 (gate Markdown) shipped their own workflow-syntax.md sections in-tree; verified those are present and consistent.
• PR feat: breadcrumb navigation, subworkflow isolation, and stop reliability #113 shipped docs/dashboard-deep-links.md in-tree; not duplicated here.

Test plan

• uv run conductor validate examples/*.yaml — all 13 examples validate cleanly.
• Markdown spot-check of new sections via diff review.
• No code changes, so no lint/typecheck/test impact.