Epic: FlameGraph/SourceCodeView improvements

Noticed some stuff during our hackathon, leaving notes here. **Features** - Global heat palette often doesn't work well. e.g. when you open a file with 3ms - 30s traces they're all the same color if globally the largest block is 30min. Per-file heat palette would be better here, but would also exaggerate the importance of even the 30s blocks. Probably no size fits all, so maybe a toggle to allow either mode? - Pick better colors for heat palette and gutters in code editor; red text on dark green background doesnt exactly scream for the attention that it deserves. Need some UX/design love here. - We should use https://github.com/lezer-parser/import-tree-sitter to port over a much richer and maintained Golang grammar for CodeMirror 6 (Lezer): https://github.com/tree-sitter/tree-sitter-go instead of using the legacy streaming parser mode: https://github.com/codemirror/legacy-modes. We built and use [LogQL](https://www.npmjs.com/package/@grafana/lezer-logql) and [PromQL](https://www.npmjs.com/package/@prometheus-io/lezer-promql) grammars for Lezer. - Toggles for Table, Flame graph, and Source code don't look very much like toggles (imo) - Maybe better alignment of gutter values. when right-aligned, `min` `ns` and `s` don't align very well. we can use `m` instead of `min`, for example. **Performance** - The mini-map should be rewritten using `<canvas>`. It was hacked together late at night from DOM nodes after an attempt to construct it using a single CSS gradient took too long to fix logic bugs with color-stop gen. - See if there's a benefit to configuring a static `gutter.markers` instead of dynamic `gutter.lineMarker()` callback that continuously creates/destroys DOM nodes: https://codemirror.net/examples/gutter/ - Stop using `DataFrameView` and `getDisplayProcessor` when processing thousands of points, both of these abstractions are very slow for the small conveniences they provide. Prefer to use `DataFrame` field values directly (by looping over `field.values.toArra…