Frame.ReadCsv("sample.csv", typeResolver = fun col ->
    match col with
    | c when c.EndsWith("Id")   -> Some "guid"
    | c when c.EndsWith("Date") -> Some "date"
    | c when c.StartsWith("Is") -> Some "bool"
    | _                         -> None)
```

The resolver is called with each column header name after the CSV has been loaded. Returning `Some "typeName"` (e.g. `"int"`, `"float"`, `"string"`, `"bool"`, `"date"`, `"guid"`, `"decimal"`, `"int64"`, `"timespan"` — same names accepted by the `schema` parameter) pins that column to the given type. Returning `None` lets Deedle infer the type as normal.

## API surface

- Added `?typeResolver: string -> string option` to:
  - `Frame.ReadCsv(path, ...)`
  - `Frame.ReadCsv(path, indexCol, ...)`
  - `Frame.ReadCsv(stream, ...)`
  - `Frame.ReadCsv(reader, ...)`
  - `Frame.ReadCsvString(csvString, ...)`
- The C# `[(Optional)]` overloads are unchanged (F# function types are not idiomatic as optional C# parameters; C# callers can still use the `schema` string parameter).

## Interaction with `schema`

When both `typeResolver` and `schema` are provided, the explicit `schema` takes precedence for any column it names. The implementation deduplicates entries so the combined schema string never exceeds the column count (which would otherwise trigger a parse error in the underlying `InferColumnTypes`).

## Implementation

- **`FrameUtils.fs`**: `readCsv` and `readString` gain a final `(string -> string option) option` parameter. Inside `readCsv`, after `CsvFile.Load`, the resolver is applied to produce `ColName=Type` named-override entries, which are merged with the explicit `schema` string and passed to the existing `InferColumnTypes` machinery.
- **`FrameExtensions.fs`**: All F# optional-parameter `ReadCsv`/`ReadCsvString` overloads forward the new parameter; C# overloads pass `None`.

## Test Status

```
dotnet test tests/Deedle.Tests/Deedle.Tests.fsproj -c Release
Passed! – Failed: 0, Passed: 657, Skipped: 0, Total: 657