Add an output pipeline that deobfuscated secrets

[jhrozek]

• Add an output pipeline - Adds an output pipeline with its own context and a possibility to buffer in case a step tells the pipeline to not forward the chunk. The buffer is stored into in case a chunk is not evaluated which means that subsequent steps can keep on buffering potentially redacted text for un-redaction later. The remainder of the buffer is flushed upon consuming the stream.
• Split out parts of the CodegateSecrets pipeline step into a secret manager - We'll want to decrypt secrets on the way from the LLM. For that we need to reuse parts of functionality that were so far in the secret encryption step. This commit splits them into a secrets manager.
• Implement secret unredaction step
• Plug in the secrets manager and output pipeline
• Clean up the secure pipeline after the completion is concluded
• Unit test the de-obfuscation of secrets

Fixes: #160
Related: #90

[jhrozek]

Here's how it looks in action:

Meanwhile in codegate logs:

Protected text:


```py app.py (5-11)
@app.route('/')
def hello():
    GITHUB_TOKEN="REDACTED<$OFjLhYcNI3ZtQi/ZH/Ss422cQhrtiHGJLnxlQ6LQUGAqEfCfdOOKq0mVadEViKLPC2M9Op+3Vnj492Ax6mxM6SBn/P2A9G3ys0hGavwg5Q==>"
    if __name__ == '__main__':
        app.run()
    return "Hello, World!"

[lukehinds]

Nice! thanks @jhrozek !